US20130254787A1 - Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising - Google Patents

Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising Download PDF

Info

Publication number
US20130254787A1
US20130254787A1 US13/663,780 US201213663780A US2013254787A1 US 20130254787 A1 US20130254787 A1 US 20130254787A1 US 201213663780 A US201213663780 A US 201213663780A US 2013254787 A1 US2013254787 A1 US 2013254787A1
Authority
US
United States
Prior art keywords
user
users
asset
information
set forth
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Abandoned
Application number
US13/663,780
Inventor
Earl Cox
Patrick M. Sheehan
Alden Lloyd Peterson
Michael Kouritzin
Surrey Kim
Jarett Hailes
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Invidi Technologies Corp
Original Assignee
Invidi Technologies Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from US11/743,544 external-priority patent/US7698236B2/en
Priority claimed from US11/944,078 external-priority patent/US20090133058A1/en
Application filed by Invidi Technologies Corp filed Critical Invidi Technologies Corp
Priority to US13/663,780 priority Critical patent/US20130254787A1/en
Assigned to INVIDI TECHNOLOGIES CORPORATION reassignment INVIDI TECHNOLOGIES CORPORATION ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: KOURITZIN, MICHAEL, PETERSON, ALDEN LLOYD, SHEEHAN, PATRICK M., HAILES, JARETT, KIM, SURREY
Publication of US20130254787A1 publication Critical patent/US20130254787A1/en
Priority to US14/949,442 priority patent/US9693086B2/en
Abandoned legal-status Critical Current

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/266Channel or content management, e.g. generation and management of keys and entitlement messages in a conditional access system, merging a VOD unicast channel into a multicast channel
    • H04N21/2668Creating a channel for a dedicated end-user group, e.g. insertion of targeted commercials based on end-user profiles
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/43Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/442Monitoring of processes or resources, e.g. detecting the failure of a recording device, monitoring the downstream bandwidth, the number of times a movie has been viewed, the storage space available from the internal hard disk
    • H04N21/44213Monitoring of end-user related data
    • H04N21/44222Analytics of user selections, e.g. selection of programs or purchase activity
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04HBROADCAST COMMUNICATION
    • H04H20/00Arrangements for broadcast or for distribution combined with broadcast
    • H04H20/10Arrangements for replacing or switching information during the broadcast or the distribution
    • H04H20/106Receiver-side switching
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04HBROADCAST COMMUNICATION
    • H04H60/00Arrangements for broadcast applications with a direct linking to broadcast information or broadcast space-time; Broadcast-related systems
    • H04H60/35Arrangements for identifying or recognising characteristics with a direct linkage to broadcast information or to broadcast space-time, e.g. for identifying broadcast stations or for identifying users
    • H04H60/45Arrangements for identifying or recognising characteristics with a direct linkage to broadcast information or to broadcast space-time, e.g. for identifying broadcast stations or for identifying users for identifying users
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/251Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/252Processing of multiple end-users' preferences to derive collaborative data
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/258Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data
    • H04N21/25866Management of end-user data
    • H04N21/25883Management of end-user data being end-user demographical data, e.g. age, family status or address
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4661Deriving a combined profile for a plurality of end-users of the same client, e.g. for family members within a home
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4662Learning process for intelligent management, e.g. learning user preferences for recommending movies characterized by learning algorithms
    • H04N21/4665Learning process for intelligent management, e.g. learning user preferences for recommending movies characterized by learning algorithms involving classification methods, e.g. Decision trees
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4667Processing of monitored end-user data, e.g. trend analysis based on the log file of viewer selections
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/80Generation or processing of content or additional data by content creator independently of the distribution process; Content per se
    • H04N21/81Monomedia components thereof
    • H04N21/812Monomedia components thereof involving advertisement data
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04HBROADCAST COMMUNICATION
    • H04H60/00Arrangements for broadcast applications with a direct linking to broadcast information or broadcast space-time; Broadcast-related systems
    • H04H60/61Arrangements for services using the result of monitoring, identification or recognition covered by groups H04H60/29-H04H60/54
    • H04H60/63Arrangements for services using the result of monitoring, identification or recognition covered by groups H04H60/29-H04H60/54 for services of sales

Definitions

  • 60/746,244 entitled: “METHOD AND APPARATUS TO PERFORM REAL-TIME ESTIMATION AND COMMERCIAL SELECTION SUITABLE FOR TARGETED ADVERTISING,” filed on May 2, 2006 and U.S. Provisional Patent Application No. 60/746,245, entitled: “METHOD AND A SYSTEM FOR DISCOVERING THE INDIVIDUAL VIEWERS OF A TELEVISION AUDIENCE (ALONG WITH THEIR ATTRIBUTES, AND BEHAVIOR) AND USING THIS INFORMATION TO ACCURATELY SELECT AND INSERT CANDIDATE ADS IN REAL-TIME FOR ACTIVELY WATCHING VIEWERS,” filed on May 2, 2006. The content of all of these applications is incorporated herein by reference as if set forth in full.
  • the present invention relates generally to targeted delivery of assets, such as advertisements or other content, in a communications network.
  • the invention relates to identifying a current network user and matching assets to the user.
  • Broadcast network content or programming is commonly provided in conjunction with associated informational content or assets. These assets include advertisements, associated programming, public-service announcements, ad tags, trailers, weather or emergency notifications and a variety of other content, including paid and unpaid content.
  • asset providers e.g., advertisers
  • information e.g., advertisements or “ads”
  • advertisers may provide ad content to a network operator such that the ad content may be interleaved with broadcast network programming during one or more programming breaks.
  • the delivery of such paid assets often subsidizes or covers the costs of the programming provided by the broadcast network. This may reduce or eliminate costs borne by the users of the broadcast network programming.
  • asset providers In order to achieve a better return on their investment, asset providers often try to target their assets to a selected audience that is deemed likely to be interested in the goods or services of the asset provider.
  • the case of advertisers on a cable television network is illustrative. For instance, an advertiser or a cable television network may target its ads to certain demographic groups based on, for example, geographic location, gender, age, income etc. Accordingly, once an advertiser has created an ad that is targeted to a desired group of viewers (e.g., targeted group) the advertiser may attempt to procure insertion times in the network programming when the targeted group is expected to be among the audience of the network programming.
  • asset targeting can be enhanced by identifying a current network user, e.g., determining demographic or other classification parameters of a putative current network user or users. This would ideally allow an asset targeting system to distinguish between different potential users of a single household, as well as identifying unknown users, such that appropriate targeting of assets can be executed.
  • the present invention enables such functionality in the context of asset delivery in communications networks, including cable television networks. Moreover, such functionality can be executed transparently, from the perspective of the network user, based on monitoring ordinary network usage activities, for example, as indicated by a click stream of a remote control. Moreover, the present invention allows such functionality to be implemented in substantially real-time, using limited processing resources.
  • the user identification functionality can be executed by an application running on a conventional digital set top box.
  • the invention also provides a mechanism for signaling the network in relation to the user identification process, for example, to enhance selection of assets for insertion into network content streams or to report information for evaluating size and composition of the audience actually reached.
  • One utility includes operating a processor to progressively incorporate, over time, a plurality of user inputs by one or more users of a user equipment device into a model of a user composition of the one or more users made up of a plurality of user classification parameters; filtering the user composition model to obtain an estimate of a current user composition of the user equipment device (e.g., where the current user composition may be made up of plurality of users); and targeting one or more assets in the communications network using the estimated current user composition.
  • a processor to progressively incorporate, over time, a plurality of user inputs by one or more users of a user equipment device into a model of a user composition of the one or more users made up of a plurality of user classification parameters; filtering the user composition model to obtain an estimate of a current user composition of the user equipment device (e.g., where the current user composition may be made up of plurality of users); and targeting one or more assets in the communications network using the estimated current user composition.
  • a current user composition of the user equipment device e.g
  • the filtered user composition model may be free of user equipment device usage patterns for the users obtained before the progressive incorporation of the plurality of user inputs.
  • the disclosed utilities can advantageously estimate a current user composition in real time in a manner substantially free of considering previously stored information (e.g., historical periodicity data for one more users).
  • a current user of a communications network can be identified without requiring persistent storage of user profiles.
  • the utility may include receiving first user inputs at a first time and then receiving second user inputs at a second time after the first time, where the second user inputs are incorporated into the model to a greater degree than are the first user inputs.
  • a reference event e.g., disconnection of power to the user equipment device
  • the progressively incorporated inputs occurred at times after the reference event (e.g., and not before the reference event).
  • the utility may advantageously incorporate a certain degree of “forgetfulness” therein so as to allow the utility to adapt to change (e.g., from a first user to multiple users to a second user), identify both known and unknown users, limit the amount of viewing information that is available in the system at any one time to address privacy concerns, and the like.
  • the utility can be adapted to quickly converge on classification parameters based on contemporaneous user inputs such that errors due to user changes are reduced.
  • An associated utility involves developing a model of a network user based on user inputs free from persistent storage of a profile of the user and using the model of the network user in targeting assets to the user. For instance, recent user inputs may be analyzed using machine learning (e.g., involving fuzzy logic) to determine classification parameters of a current user.
  • the targeting of assets may include receiving one or more lists of assets for delivery at the user equipment device; obtaining one or more targeting parameters for the one or more lists of assets; and determining a level of correspondence between the user classification parameters of the current user composition of the user equipment device and the targeting parameters for the one or more lists of assets. For instance, the determining may be performed in multiple dimensions relating to multiple classification and targeting parameters.
  • the utility may include voting for at least one asset of the one or more lists of assets based on the determined level of correspondence between the user classification parameters of the current user composition of the user equipment device and the targeting parameters for the one or more lists of assets. For instance, the at least one voted-for asset may include targeting parameters having a higher level of correspondence with the user classification parameters of the current user composition that does a non-voted for asset.
  • the voting may include sending, from the user equipment device, an indication at least one asset of the lists of assets having targeting parameters with a higher level of correspondence with the user classification parameters of the current user composition that does at least one other asset of the lists of assets.
  • the utility may further include receiving the one or more lists of assets for delivery at a plurality of additional user equipment devices; determining a level of correspondence between user classification parameters of one or more current user compositions of the plurality of additional user equipment devices and the targeting parameters for the one or more lists of assets; voting for at least one asset of the one or more lists of assets based on the determined level of correspondence between the user classification parameters of the current user compositions of the plurality of additional user equipment devices and the targeting parameters for the one or more lists of assets; and aggregating the votes of the user equipment device and the plurality of additional user equipment devices.
  • the utility may further include receiving one or more updated lists of assets for delivery at the user equipment device and the plurality of additional user equipment devices.
  • the utility may include progressively incorporating, over time, a plurality of user inputs by users of a plurality of additional user equipment devices into a plurality of models of user composition of the users made up of a plurality of user classification parameters; filtering the user composition models to obtain estimates of current user compositions of the plurality of additional user equipment devices; and aggregating the current user compositions of the user equipment device and the plurality of additional user equipment devices to obtain a current user composition of an aggregated audience, where the targeting includes using the aggregated audience current user composition for use in targeting one or more assets in the communications network. For instance, a level of correspondence between the user classification parameters of the aggregated audience current user composition and the targeting parameters may be determined for the one or more lists of assets.
  • the user composition model may be filtered to obtain current user composition estimates in various manners.
  • certain aspects of the invention relate to processing corrupted, distorted and/or partial data observations received from the measurement device to infer information about a signal of interest, such as a substantially real time estimate of the state of the signal (e.g., current user composition) at a time of interest.
  • a filter system can provide practical approximations of optimized nonlinear filter solutions based on certain constraints on allowable states or combinations therefore inferred from the observation environment.
  • a method and apparatus for developing an observation model with respect to data or measurements obtained from the device under analysis.
  • the system models the input measurements as a Markov chain, whose transitions depend upon the signal.
  • the observation model may take into account exogenous information or information external to (though not necessarily independent of) the input measurements.
  • the input measurements reflect a click stream of DSTB.
  • the click stream may reflect channel selection events and/or other inputs, e.g., related to volume control.
  • the observation model may further involve programming information (e.g., downloaded from a network platform such as a Head End) associated with selected channels. In this case, it is the click stream information that is processed as a Markov chain.
  • Desired information related to the device can then be obtained by estimating the state of the signal at a time of interest.
  • the signal may represent a user composition (involving one or more users and/or associated demographics) and an additional factor affecting the click stream such as a channel changing regime as discussed in more detail below.
  • a state of the signal at a past, present or future time can be determined, e.g., to provide user composition information for use in connection with an asset targeting system.
  • a system generates substantially real time estimates of the probability distribution for a signal state based on both the observations and an observation signal model.
  • a nonlinear filter system can be used to provide an estimate of the signal based on the observation model.
  • the nonlinear filter system may involve a nonlinear filter model and an approximation filter for approximating an optimal nonlinear filter solution.
  • the approximation filter may include a particle filter or a discrete state filter for enabling substantially real time estimates of the signal based on the observation model.
  • the nonlinear filter system allows for estimates that incorporate user compositions including more than one viewer and adapting to changes in the potential audience, e.g., additions of previously unknown persons or departures of prior users with respect to the potential audience.
  • a system uses an estimate obtained by applying a filter, with its associated signal and observation models, to a sequence of observations to obtain information of interest with respect to the signal. Specifically, information for a past, present or future time can be obtained based on an estimated probability distribution of the signal at the time of interest. In the case of analyzing usage of a DSTB, the identity and/or demographics of a user or users of the DSTB at a particular time can be determined from the signal state.
  • This information may be used, for example, to “vote” or identify appropriate assets for an upcoming commercial or programming spot, to select an asset from among asset options for delivery at the DSTB and/or to determine or report a goodness of fit of a delivered asset with respect to the user or users who received the asset.
  • a system for use in targeting assets to users of user equipment devices in a communications network, for example, a cable television network.
  • the system involves: developing an observation model based on inputs (e.g., click stream data) by one or more users with respect to a user equipment device (e.g., a DSTB); modeling the signal as reflective of at least a user composition of one or more users of said user equipment device with respect to time; determining the likelihood of various user compositions at a time of interest among possible states of the signal; and using the estimated user composition in targeting an asset for the user equipment device.
  • filtering theory is applied with respect to inputs, such as a click stream, of a user equipment device so as to yield an estimate indicative of user composition.
  • the observations can be modeled as a Markov chain.
  • the model of the signal allows for representation of the user composition as including two or more users. Accordingly, multiple user situations can be identified for use in targeting assets and/or better evaluating audience size and composition (e.g., to improve valuation and billing for asset delivery).
  • the signal model preferably allows for representation of a change in user composition, e.g., addition or removal of a person from a user audience.
  • a nonlinear filter may be defined to estimate the signal based on the observation model.
  • the signal may model the user composition of a household with respect to time and audience classification parameters (e.g., demographics of one or more current users) can be estimated as a function of the state of the signal at a time of interest.
  • audience classification parameters e.g., demographics of one or more current users
  • an approximation filter may be provided for approximating the operation of the nonlinear filter.
  • the approximation filter may include a particle filter or a discrete space filter as described below.
  • the approximation filter may implement at least one constraint with respect to one or more signal components.
  • the constraint may operate to treat one component of the signal as invariant with respect to a time period where a second component is allowed to vary. Moreover, the constraint may operate to treat at least one state of a first component as illegitimate or to treat some combination of states of different signal components as illegitimate. For example, in the case of a click stream of a DSTB, the occurrence of a click event indicates the certain presence of at least one person. Accordingly, only user compositions corresponding to the presence of at least one person are permissible at the time of a click event. Other permissible or impermissible combinations may relate incomes to locations.
  • the constraints may be implemented in connection with a finite space approximation filter.
  • values incident on an illegitimate cell may be repositioned, e.g., proportionately moved to neighboring legitimate cells.
  • the approximation filter can quickly converge on a legitimate solution without requiring undue processing resources.
  • the constraint operates to define at least one potential calculated state as illegitimate
  • the approximation filter may redistribute one or more counts associated therewith.
  • the approximation filter may be operative to inhibit convergence on an illegitimate state.
  • the approximation filter is designed to avoid convergence on a user composition for a DSTB that is logically impossible or unlikely (a click event when no user is present) or deemed illegitimate by rule (an income range not permitted for a given location). In one implementation, this is accomplished by adding seed counts to legitimate cells of a discrete space filter to inhibit convergence with respect to an illegitimate cell.
  • the user composition information is processed at the DSTB. That is, user information is processed at the DSTB and used for voting, asset selection and/or reporting.
  • click stream data may be directed to a separate platform, such as a Head End, where the user composition information can be estimated, e.g., where messaging bandwidth is sufficient and DSTB processing resources are limited.
  • the user composition information (as opposed to, e.g., asset vote information) may be transmitted to a Head End or other platform for use in selecting content for insertion.
  • the estimated user composition information may be used by an asset targeting system.
  • the information may be provided to a network platform such as a Head End that is operative to insert assets into a content stream of the network.
  • the platform may utilize inputs from multiple DSTBs to select assets for insertion into available network bandwidth. Additional information, such as information reflecting the per user value of asset delivery, may be utilized in this regard.
  • the platform may process information from multiple user equipment devices as an observation model and apply an appropriately configured filter with respect to the observation model to estimate an overall composition of a network audience at a time of interest.
  • stochastic control theory is applied to the problem of asset selection, e.g., selecting the optimal set of commercial assets to communicate through a limited number of advertising insertion channels.
  • stochastic control theory has been applied in contexts where the state of a system is randomly (time) varying and possibly the exact consequences of various controls applied to the system are only known probabilistically.
  • sampled viewer estimates from DSTBs received at the Head End are taken to be observations of the system of probability distributions over household viewing states, of arriving advertising contracts, and of ad sale and delivery, in order to allow control decisions regarding which contracts with advertisers to accept.
  • Stochastic control is used to optimize some utility function of the system, e.g., stable profitability.
  • Another manner of filtering the user composition model to obtain the current user composition estimates may involve utilizing machine learning (e.g., fuzzy logic including fuzzy sets and/or fuzzy rules) as part of matching assets to current users.
  • the machine learning system may involve identifying classification parameters of at least one (e.g., current) user based on evidence aggregated from user inputs collected in a learning mode. These inputs may, for example, be analyzed based on correlated programming information, or based on programming independent characteristics, e.g., volume settings or quickness of the click process.
  • fuzzy logic may be used to identify a number of discrete users in an audience (e.g., number of members of a household) and/or to determine one or more classification parameters of a user or users. This may be based on user inputs such as a click stream of a remote control. Thus, user inputs may be monitored and associated with values related to the classification parameter(s). These values can then be treated as points in a fuzzy set within a model of a plurality of user classification parameters. In one implementation, a number of user inputs may be monitored to aggregate points in a fuzzy set.
  • the current user composition estimation and/or matching may involve multiple dimensions related to multiple classification parameters (e.g., age, gender, income, etc.), and the aggregated points may be used to define one or more features of a multidimensional feature terrain.
  • the feature terrain may be processed to remove noise and to reduce the set of gradients in the terrain, for example, by clustering features.
  • the remaining features of the processed feature terrain can then be used to identify each user in an audience and determine one or more classification parameters for each user. Similar processing can be used to identify viewing patterns as a function of time (periodicity). For example, different terrains can be developed for different time periods, e.g., different times of day.
  • fuzzy logic may be used to develop a characterization of a target audience of a network programming event.
  • the target audience of an asset may be defined by a demographic profile including a number of demographic parameter values. These values may be associated with a series of fuzzy numbers or fuzzy sets.
  • An additional implementation of fuzzy logic may be used to correlate the fuzzy numbers with classification parameters of putative user. For example, a congruent similarity function may be used to match the audience characterization or targeting parameters to the classification parameters. Similar processing can be used to match a periodicity pattern to an identified user. Alternatively, where different terrains are developed for different times, as noted above, such periodicity is reflected in the terrains; that is, time becomes a dimension of the terrain set.
  • a match may be determined based on a combination of the degree of correlation of the user classification parameters to the ad targeting parameters and the likelihood that an appropriate viewer will be watching at the time of the ad delivery.
  • the resulting match may be used to “vote” for assets to be inserted into content streams of the network to select ads for delivery and/or to report a “goodness of fit” of a user receiving the asset to the asset targeting parameters.
  • the noted utility may also be operative to determine whether the user equipment device is “on” and to determine whether any user is present at the user equipment device.
  • the user inputs or click stream data can be processed at the user equipment device or at another location, e.g., raw or preprocessed click stream data may be transmitted to a head end for processing to determine classification parameter information. For example, this may be done where messaging bandwidth is sufficient and user equipment device resources are limited.
  • the machine learning system may be a substantially unsupervised system. That is, the system can accumulate evidence and thereby learn a composition of a user set, such as a viewing audience, without requiring a training process in which the system is provided knowledge about or examples of usage (e.g., viewing) patterns. In this manner, the system can readily adapt to changes, e.g., changes in the viewing audience or viewing audience demographics due to, for example, additions to or departures from the household, changing demographics due to aging, change of income, etc., addition of a television set (e.g., in a child's room) that impacts viewership, etc. Moreover, the system can operate substantially autonomously, thereby substantially avoiding the need for any supervised set-up or retraining process.
  • functionality for identifying a user can be executed at a user equipment device. It has been recognized that a current user can be effectively identified based on analysis of user inputs at a user equipment device.
  • An associated utility in accordance with the present invention involves receiving user inputs at the user equipment device and analyzing the inputs to associate audience classification parameters with the user using a machine learning system.
  • the inputs may relate to a click stream of a remote control device reflecting program selections, volume control inputs and the like.
  • the machine learning system is preferably capable of learning in a substantially unsupervised fashion. Fuzzy logic can be used to analyze these inputs on an individual basis to obtain evidence concerning the classification parameters of the user. This evidence can then be aggregated and analyzed using fuzzy logic to determine classification parameters of a user.
  • a user equipment device is operative to signal a broadcast network regarding a user of the device.
  • An associated utility involves determining, at the user equipment device, user information regarding the user of the device based at least in part on user inputs to the device, and signaling the broadcast network based on user information.
  • the user information may include classification parameters of the user.
  • the signals transmitted to the broadcast network may reflect the results of a matching process whereby user classification information is compared to targeting information for an asset.
  • the information transmitted across the network need not include any classification information regarding the user.
  • Such signaling information may be used, for example, to vote for assets to be inserted into network content streams or to report information regarding assets actually delivered at the user equipment device, e.g., for measuring the size and/or composition of the audience.
  • FIG. 1 illustrates delivery of assets to different users watching the same programming channel.
  • FIG. 2 illustrates audience aggregation across multiple programming networks.
  • FIG. 3 illustrates a virtual channel in the context of audience aggregation.
  • FIG. 4 illustrates targeted asset insertion being implemented at Customer Premises Equipment (CPEs).
  • CPEs Customer Premises Equipment
  • FIG. 5 illustrates asset options being transmitted from a headend on separate asset channels.
  • FIG. 6 illustrates a messaging sequence between a CPE, a network platform, and a traffic and billing (T&B) system.
  • T&B traffic and billing
  • FIG. 7A illustrates an example of CPEs that include a television set and a Digital Set Top Box (DSTB) as used by a plurality of users.
  • DSTB Digital Set Top Box
  • FIG. 7B illustrates a user classifier
  • FIG. 8 is a flow chart illustrating a process for implementing time-slot and targeted impression buys.
  • FIG. 9 illustrates an overview of a classifier process in accordance with the present invention.
  • FIG. 10 is a stay transition graph illustrating a process for handling click stream data in accordance with the present invention.
  • FIGS. 11-21 illustrate learning mode operation of the classifier in accordance with the present invention.
  • FIGS. 22-26 illustrate working mode operation of the classifier in accordance with the present invention.
  • FIG. 27 is a block diagram illustrating the basic functional components of a classifier in accordance with the present invention.
  • FIG. 28 is a schematic diagram of a targeted advertising system in accordance with one embodiment.
  • FIG. 29 illustrates the REST structure in accordance with one embodiment.
  • FIG. 30 illustrates a cell structure for a cell of a discrete space filter in accordance with one embodiment.
  • FIG. 31 is a flowchart illustrating a filter evolution process in accordance with one embodiment.
  • FIG. 32 is a block diagram illustrating a process for simulating events in accordance with one embodiment.
  • the present invention relates to various structure and functionality for delivery of targeted assets, classification of network users, matching of asset targeting parameters to audience classification parameters and network monitoring for use in a communications network.
  • the invention has particular application with respect to networks where content is broadcast to network users.
  • content may be broadcast in a variety of networks including, for example, cable and satellite television networks, satellite radio networks, IP networks used for multicasting content and networks used for podcasts or telephony broadcasts/multicasts.
  • Content may also be broadcast over the airwaves though, as will be understood from the description below, certain aspects of the invention make use of bi-directional communication channels which are not readily available, for example, in connection with conventional airwave based televisions or radios (i.e., such communication would involve supplemental communication systems).
  • the content may be consumed in real time or stored for subsequent consumption.
  • the invention is not limited to such contexts but, rather, has application to a variety of networks and transmission modes.
  • the targeted assets may include any type of asset that is desired to be targeted to network users. It is noted that such targeted assets are sometimes referred to as “addressable” assets (though, as will be understood from the description below, targeting can be accomplished without addressing in a point-to-point sense).
  • these targeted assets may include advertisements, internal marketing (e.g., information about network promotions, scheduling or upcoming events), public service announcements, weather or emergency information, or programming.
  • the targeted assets may be independent or included in a content stream with other assets such as untargeted network programming. In the latter case, the targeted assets may be interspersed with untargeted programming (e.g., provided during programming breaks) or may otherwise be combined with the programming as by being superimposed on a screen portion in the case of video programming.
  • the present invention relates to identifying members of an audience, determining classification information for those users, determining which user or users may be watching at a time of interest, and matching assets to the identified audience.
  • the matching related functionality is useful in a variety of contexts in a targeted asset delivery system. Accordingly, an overview of the targeted asset delivery system is first provided below. Thereafter, the matching related functionality and associated structure is described in detail.
  • the inventive system allows for delivery of targeted assets such as advertising for use in addressing certain shortcomings or inefficiencies of conventional broadcast networks.
  • targeted assets such as advertising for use in addressing certain shortcomings or inefficiencies of conventional broadcast networks.
  • targeting entails delivering assets to desired groups of individuals or individuals having desired characteristics. These characteristics or audience classification parameters may be defined based on personal information, demographic information, psychographic information, geographic information, or any other information that may be relevant to an asset provider in identifying a target audience.
  • targeting is program independent in recognition that programming is a highly imperfect mechanism for targeting of assets.
  • targeted asset delivery would allow for targeting with a range of granularities including very fine granularities. For example, it may be desired to target a group, such as based on a geographical grouping, a household characterization or even an individual user characterization.
  • the present invention accommodates program independent targeting, targeting with a high degree of granularity and targeting based on a variety of different audience classifications.
  • FIGS. 1 and 2 illustrate two different contexts of targeted asset delivery supported in accordance with the present invention.
  • FIG. 1 illustrates the delivery of different assets, in this case ads, to different users watching the same programming channel, which may be referred to as spot optimization.
  • three different users 500 - 502 are depicted as watching the same programming, in this case, denoted “Movie of the Week.”
  • the users 500 - 502 each receive a different asset package.
  • user 500 receives a digital music player ad and a movie promo
  • user 501 receives a luxury car ad and a health insurance ad
  • user 502 receives a minivan ad and a department store ad.
  • a single asset provider e.g., a motor vehicle company
  • may purchase a spot and then provide different asset options for the spot e.g., sports car, minivans, pickup trucks, etc.
  • asset options e.g., sports car, minivans, pickup trucks, etc.
  • separate advertisers may collectively purchase a spot and then provide ads for their respective products (e.g., where the target audiences of the advertisers are complementary).
  • these different asset packages may be targeted to different audience demographics. In this manner, assets are better tailored to particular viewers of a given program who may fall into different demographic groups.
  • spot optimization refers to the delivery of different assets (by one or multiple asset providers) in a given spot.
  • FIG. 2 illustrates a different context of the present invention, which may be termed audience aggregation.
  • three different users 600 - 602 viewing different programs associated with different channels may receive the same asset or asset package.
  • each of the users 600 - 602 receives a package including a digital music player ad and a movie promo in connection with breaks associated with their respective channels.
  • the users 600 - 602 are shown as receiving the same asset package for purposes of illustration, it is likely that different users will receive different combinations of assets due to differences in classification parameters.
  • users over multiple channels can be aggregated (relative to a given asset and time window) to define a virtual channel having significant user numbers matching a targeted audience classification.
  • audience aggregation allows for the possibility of aggregating users over a number of low share channels to define a significant asset delivery opportunity, perhaps on the order of that associated with one of the high share networks. This can be accomplished, in accordance with the present invention, using equipment already at a user's premises (i.e., an existing CPE). Such a virtual channel is graphically illustrated in FIG. 3 , though this illustration is not based on actual numbers.
  • audience aggregation refers to the delivery of the same asset in different spots to define an aggregated audience. These different spots may occur within a time window corresponding to overlapping (conflicting) programs on different channels. In this manner, it is likely that these spots, even if at different times within the window, will not be received by the same users.
  • targeted asset insertion can be implemented at the CPEs. This may involve a forward-and-store functionality.
  • the CPE 800 receives a programming stream 802 and an asset delivery stream 804 from the headend 808 . These streams 802 and 804 may be provided via a common signal link such as a coaxial cable or via separate communications links.
  • the asset delivery stream 804 may be transmitted to the CPE 800 via a designated segment, e.g., a dedicated frequency range, of the available bandwidth or via a programming channel that is opportunistically available for asset delivery, e.g., when it is otherwise off air.
  • the asset delivery stream 804 may be provided on a continuous or intermittent basis and may be provided concurrently with the programming stream 802 .
  • the programming stream 802 is processed by a program-decoding unit, such as DSTB, and programming is displayed on television set 814 .
  • the programming stream 802 may be stored in programming storage 815 for CPE insertion.
  • the asset is stored in a designated storage space 806 of the CPE 800 .
  • substantial storage at the CPE 800 may be required in this regard.
  • DVR digital video recorder
  • a selector 810 is implemented as a processor running logic on the CPE 800 .
  • the selector 810 functions analogously to the headend selector described above to identify breaks 816 and insert appropriate assets.
  • the assets may be selected based on classification parameters of the household or, more preferably, a user within the household.
  • Such information may be stored at the CPE 800 or may be determined based on an analysis of viewing habits such as a click stream from a remote control as will be described in more detail below. Certain aspects of the present invention can be implemented in such a CPE insertion environment.
  • FIG. 5 a different architecture is employed. Specifically, in FIG. 5 , asset options transmitted from headend 910 synchronously with a given break on a given channel for which targeted asset options are supported.
  • the CPE 900 includes a channel selector 902 , which is operative to switch to an asset channel associated with a desired asset at the beginning of a break and to return to the programming channel at the end of the break.
  • the channel selector 902 may hop between channels (between asset channels or between an asset channel and the programming channel) during a break to select the most appropriate assets.
  • logic resident on the CPE 900 controls such hopping to avoid switching to a channel where an asset is already in progress. As described below, this logic can be readily implemented, as the schedule of assets on each asset channel is known.
  • all of this is implemented invisibly from the perspective of the user of set 904 .
  • the different options may be provided, at least in part, in connection with asset channels 906 or other bandwidth segments (separate from programming channels 908 ) dedicated for use in providing such options.
  • certain asset options may be inserted into the current programming channel 908 .
  • the architecture of FIG. 5 has the advantage of not requiring substantial storage resources at the CPE 900 such that it can be immediately implemented on a wide scale basis using equipment that is already in the field.
  • the determination of which asset to show may be made at the headend. For example, an asset may be selected based on voting as described below, and inserted at the headend into the programming channel without options on other asset channels. This would achieve a degree of targeting but without spot optimization opportunities as described above. Still further, options may be provided on other asset channels, but the selection as between those channels may be determined by the headend. For example, information about a household or user (e.g., brand of car owned, magazines subscribed to, etc.) stored on the headend may be used to match an asset to a household or user. That information, which may be termed “marketing labels,” may be used by the headend to control which asset is selected by the CPE.
  • marketing labels may be used by the headend to control which asset is selected by the CPE.
  • the CPE may be instructed that it is associated with an “ACME preferred” customer.
  • ACME preferred metadata When an asset is disseminated with ACME preferred metadata, the CPE may be caused to select that asset, thereby overriding (or significantly factoring with) any other audience classification considerations.
  • such operation may entail certain concerns relating to sensitive information or may compromise audience classification based targeting in other respects.
  • the system of the present invention involves the transmission of asset options in time alignment or synchronization with other assets on a programming channel, where the asset options are at least partially provided via separate bandwidth segments, e.g. channels at least temporarily dedicated to targeted asset delivery.
  • asset options may typically be transmitted in alignment with a break in programming, it may be desired to provide options opposite continuing programming (e.g., so that only subscribers in a specified geographic area get a weather announcement, an emergency announcement, election results or other local information while others get uninterrupted programming). Selection as between the available options is implemented at the user's premises, as by a DSTB in this implementation.
  • asset options are made available for better targeting, without the requirement for substantial storage resources or equipment upgrades at the user's premises (e.g., as might be required for a forward-and-store architecture).
  • existing DSTBs can be configured to execute logic for implementing the system described below by downloading and/or preloading appropriate logic.
  • bandwidth identification e.g., identifying bandwidth that is opportunistically available in relation to a node filter
  • Efficient use of available bandwidth involves both optimizing the duty cycle or asset density of an available bandwidth segment (i.e., how much time, of the time a bandwidth segment is available for use in transmitting asset options, is the segment actually used for transmitting options) and the value of the options transmitted.
  • the former factor is addressed, among other things, by improved scheduling of targeted asset delivery on the asset channels in relation to scheduled breaks of the programming channels.
  • FIG. 6 illustrates an associated messaging sequence 1000 in this regard as between a CPE 1002 such as a DSTB, a network platform for asset insertion such as a headend 1004 and a traffic and billing (T&B) system 1006 used in the illustrated example for obtaining asset delivery orders or contracts and billing for asset delivery.
  • a CPE 1002 such as a DSTB
  • a network platform for asset insertion such as a headend 1004
  • T&B traffic and billing
  • the functionality of the T&B system 1006 may be split between multiple systems running on multiple platforms and the T&B system 1006 may be operated by the network operator or may be separately operated.
  • the illustrated sequence begins by loading contract information 1008 from the T&B system 1006 onto the headend 1004 .
  • An interface associated with system 1006 allows asset providers to execute contracts for dissemination of assets based on traditional time-slot buys (for a given program or given time on a given network) or based on a certain audience classification information (e.g., desired demographics, psychographics, geography, and/or audience size). In the latter case, the asset provider or network may identify audience classification information associated with a target audience.
  • the system 1006 uses this information to compile the contract information 1008 , which identifies the asset that is to be delivered together with delivery parameters regarding when and to whom the asset is to be delivered.
  • the illustrated headend 1004 uses the contract information together with a schedule of breaks for individual networks to compile an asset option list 1010 on a channel-by-channel and break-by-break basis. That is, the list 1010 lists the universe of asset options that are available for voting purposes for a given break on a given programming channel together with associated metadata identifying the target audience for the asset, e.g., based on audience classification information.
  • the transmitted list 1010 may encompass all supported programming channels and may be transmitted to all participating users, or the list may be limited to one or a subset of the supported channels e.g., based on an input indicating the current channel or the most likely or frequent channels used by a particular user or group of users.
  • the list 1010 is transmitted from the headend 1004 to the CPE 1002 in advance of a break for which options are listed.
  • the CPE 1002 submits a vote 1012 back to the headend 1004 . More specifically, the CPE 1002 first identifies the classification parameters for the current user(s) and perhaps the current channel being watched, identifies the assets that are available for an upcoming break (for the current channel or multiple channels) as well as the target audience for those assets and determines a “fit” of one or more of those asset options to the current classification. In one implementation, each of the assets is attributed a fit score for the user(s), e.g., based on a comparison of the audience classification parameters of the asset to the putative audience classification parameters of the current user(s).
  • the CPE 1002 issues the vote 1012 indicating the most appropriate asset(s). Any suitable information can be used to provide this indication. For example, all scores for all available asset options (for the current channel or multiple channels) may be included in the vote 1012 . Alternatively, the vote 1012 may identify a subset of one or more options selected or deselected by the CPE 1002 , with or without scoring information indicating a degree of the match and may further include channel information.
  • the headend 1004 instructs CPEs ( 1002 ) to return fit scores for the top N asset options for a given spot, where N is dynamically configurable based on any relevant factor such as network traffic levels and size of the audience.
  • this voting occurs shortly before the break at issue such that the voting more accurately reflects the current status of network users.
  • votes are only submitted for the programming channel to which the CPE is set, and votes are submitted periodically, e.g., every fifteen minutes.
  • the headend 1004 compiles votes 1012 from CPEs 1002 to determine a set of selected asset options 1014 for a given break on a supported programming channel. As will be understood from the description below, such votes 1012 may be obtained from all relevant and participating CPEs 1002 (who may be representative of a larger audience including analog or otherwise non-participating users) or a statistical sampling thereof. In addition, the headend 1004 determines the amount of bandwidth, e.g., the number of dedicated asset option channels, that is available for transmission of options in support of a given break for a given programming channel.
  • the amount of bandwidth e.g., the number of dedicated asset option channels
  • the headend 1004 assembles a flotilla of assets, e.g., the asset options having the highest vote values or the highest weighted vote values where such weighting takes into account value per user or other information beyond classification fit.
  • a flotilla may include asset options inserted on the current programming channel as well as on asset channels, though different insertion processes and components may be involved for programming channel and asset channel insertion.
  • assets may be assembled independently or largely independently of voting, for example, certain public service spots or where a certain provider has paid a premium for guaranteed delivery.
  • voting may be unnecessary (though voting may still be used to select the options).
  • the flotilla is assembled into sets of asset options for each dedicated asset channel, where the time length of each set matches the length of the break, such that channel hopping within a break is unnecessary.
  • the CPE 1002 may navigate between the asset channels to access desired assets within a break (provided that asset starts on the relevant asset channels are synchronized).
  • the flotilla matrix (where columns include options for a given spot and rows correspond to channels) need not be rectangular. Stated differently, some channels may be used to provide asset options for only a portion of the break, i.e., may be used at the start of the break for one or more spots but are not available for the entire break, or may only be used after one or more spots of a break have aired.
  • a list of the selected assets 1014 and the associated asset channels is then transmitted together with metadata identifying the target audience in the illustrated implementation. It will be appreciated that it may be unnecessary to include the metadata at this step if the CPE 1002 has retained the asset option list 1010 .
  • This list 1014 is preferably transmitted shortly in advance of transmission of the asset 1016 (which includes sets of asset options for each dedicated contact options channel used to support, at least in part, the break at issue).
  • the CPE 1002 receives the list of selected asset options 1014 and associated metadata and selects which of the available options to deliver to the user(s). For example, this may involve a comparison of the current audience classification parameter values (which may or may not be the same as those used for purposes of voting) to the metadata associated with each of the asset options.
  • the selected asset option is used to selectively switch the CPE 1002 to the corresponding dedicated asset options channel to display the selected asset 1016 at the beginning of the break at issue.
  • One of the asset option sets for example, the one comprised of the asset receiving the highest vote values, may be inserted into the programming channel so that switching is not required for many users.
  • the voting CPEs are at least somewhat representative of the universe of all users, a significant degree of targeting is thereby achieved even for analog or otherwise non-participating users.
  • the voters serve as proxies for non-voting users.
  • the CPE 1002 returns to the programming channel at the conclusion of the break. Preferably, all of this is transparent from the perspective of the user(s), i.e., preferably no user input is required.
  • the system may be designed so that any user input overrides the targeting system.
  • the change will be implemented as if the targeting system was not in effect (e.g., a command to advance to the next channel will set the CPE to the channel immediately above the current programming channel, without regard to any options currently available for that channel, regardless of the dedicated asset channel that is currently sourcing the television output).
  • the system of the present invention preferably implements a reporting system by which individual CPEs 1002 report back to the headend 1004 what asset or assets were delivered at the CPE 1002 and, optionally, to whom (in terms of audience classification). Additionally, the reports may indicate where (on what programming channel) the asset was delivered and how much (if any) of the asset was consumed.
  • Such reports 1018 may be provided by all participating CPEs 1002 or by a statistical sampling thereof. These reports 1018 may be generated on a break-by-break basis, periodically (e.g., every 15 minutes) or may be aggregated prior to transmission to the headend 1004 .
  • Reports may be transmitted soon after delivery of the assets at issue or may be accumulated, e.g., for transmission at a time of day where messaging bandwidth is more available. Moreover, such reporting may be coordinated as between the CPEs 1002 so as to spread the messaging load due to reporting.
  • the reports 1018 can be used to provide billing information 1020 to the T&B system 1006 for valuing the delivery of the various asset options.
  • the billing information 1020 can be used by the T&B system 1006 to determine how large an audience received each option and how well that audience matched the target audience.
  • a fit score may be generated for particular asset options based on a comparison of the audience classification to the target audience. This score may be on any scale, e.g., 1-100. Goodness of fit may be determined based on this raw score or based on characterization of this score such as “excellent,” “good,” etc.
  • This information may in turn be provided to the asset provider, at least in an aggregated form.
  • the network operator can bill based on guaranteed delivery of targeted messages or scale the billing rate (or increase delivery) based on goodness of fit as well as audience size.
  • the reports (and/or votes) 1018 can also provide a quick and detailed measurement of user distribution over the network that can be used to accurately gauge ratings, share, demographics of audiences and the like.
  • this information can be used to provide future audience estimation information 1022 , for example, to estimate the total target universe based on audience classification parameters.
  • the present invention allows a network operator such as an MSO to sell asset delivery under the conventional asset delivery (time-slot) buy paradigm or under the new commercial impression paradigm or both.
  • a particular MSO may choose to sell asset delivery space for the major networks (or for these networks during prime time) under the old time-slot buy paradigm while using the commercial impression paradigm to aggregate users over multiple low market share networks.
  • Another MSO may choose to retain the basic time-slot buy paradigm while accommodating asset providers who may wish to fill a given slot with multiple options targeted to different demographics.
  • Another MSO may choose to retain the basic time-slot buy paradigm during prime time across all networks while using the targeted impression paradigm to aggregate users at other times of the day.
  • the targeted impression paradigm may be used by such MSOs only for this limited purpose.
  • FIG. 8 is a flow chart illustrating an associated process 1200 .
  • An asset provider (or agent thereof) can initiate the illustrated process 1200 by accessing ( 1202 ) a contracting platform as will be described below.
  • an asset provider can work with the sales department or other personnel of a system operator or other party who accesses such a platform.
  • an automated buying system may be employed to interface with such a platform via a system-to-system interface.
  • This platform may provide a graphical user interface by which an asset provider can design a dissemination strategy and enter into a corresponding contract for dissemination of an asset.
  • the asset provider can then use the interface to select ( 1204 ) to execute either a time-slot buy strategy or a targeted impression buy strategy.
  • the asset provider can then use the user interface to specify ( 1206 ) a network and time-slot or other program parameter identifying the desired air times and frequency for delivery of the asset.
  • a network and time-slot or other program parameter identifying the desired air times and frequency for delivery of the asset.
  • an asset provider may elect to air the asset in connection with specifically identified programs believed to have an appropriate audience.
  • the asset provider may specify that the asset is to appear during the first break or during multiple breaks during the program.
  • the asset provider may further specify that the asset is to be, for example, aired during the first spot within the break, the last spot within the break or otherwise designate the specific asset delivery slot.
  • the MSO causes the asset to be embedded ( 1208 ) into the specified programming channel asset stream.
  • the asset is then available to be consumed by all users of the programming channel.
  • the MSO then bills ( 1210 ) the asset provider, typically based on associated ratings information. For example, the billing rate may be established in advance based on previous rating information for the program in question, or the best available ratings information for the particular airing of the program may be used to bill the asset provider.
  • the conventional time-slot buy paradigm is limited to delivery to all users for a particular time-slot on a particular network and does not allow for targeting of particular users of a given network or targeting users distributed over multiple networks in a single buy.
  • the asset provider can use the user interface as described in more detail below to specify ( 1212 ) audience classification and other dissemination parameters.
  • audience classification parameters the asset provider may specify the gender, age range, income range, geographical location, lifestyle interest or other information of a targeted audience.
  • the additional dissemination parameters may relate to delivery time, frequency, audience size, or any other information useful to define a target audience. Combinations of parameters may also be specified. For example, an asset provider may specify an audience size of 100,000 in a particular demographic group and further specify that the asset is not delivered to any user who has already received the asset a predetermined number of times.
  • the targeted asset system of the present invention is operative to target appropriate users. For example, this may involve targeting only selected users of a major network. Additionally or alternatively, this may involve aggregating ( 1214 ) users across multiple networks to satisfy the audience specifications. For example, selected users from multiple programming channels may receive the asset within a designated time period in order to provide an audience of the desired size, where the audience is composed of users matching the desired audience classification.
  • the user interface preferably estimates the target universe based on the audience classification and dissemination parameters such that the asset provider receives an indication of the likely audience size.
  • the aggregation system may also be used to do time of day buys.
  • an asset provider could specify audience classification parameters for a target audience and further specify a time and channel for airing of the asset. CPEs tuned to that channel can then select the asset based on the voting process as described herein.
  • asset providers may designate audience classification parameters and a run time or time range, but not the programming channel. In this manner, significant flexibility is enabled for designing a dissemination strategy. It is also possible for a network operator to disable some of these strategy options, e.g., for business reasons.
  • the targeted asset system of the present invention is operative to provide the asset as an option during one or more time-slots of one or more breaks.
  • multiple asset options may be disseminated together with information identifying the target audience so that the most appropriate asset can be delivered at individual CPEs.
  • audience aggregation the asset may be provided as an option in connection with multiple breaks on multiple programming channels.
  • the system receives and processes ( 1218 ) reports regarding actual delivery of the asset by CPEs and information indicating how well the actual audience fit the classification parameters of the target audience.
  • the asset provider can then be billed ( 1220 ) based on guaranteed delivery and goodness of fit based on actual report information.
  • PII personally identifiable information
  • any information that may entail privacy concerns or identify network usage information may be considered sensitive information.
  • the system learns of current network conditions prior to transmission of asset options via votes that identify assets without any sensitive information.
  • Reports may also be limited to identifying assets that have been delivered (which assets are associated with target audience parameters) or characterization of the fit of audience classification parameters of a user(s) to a target audience definition. Even if it is desired to associate reports with particular users, e.g., to account for ad skipping as discussed below, such association may be based on an identification code or address not including PII. In any event, identification codes or any other information deemed sensitive can be immediately stripped and discarded or hashed, and audience classification information can be used only in anonymous and aggregated form to address any privacy concerns.
  • sensitive information such as a MAC or IP address (which may be included in a designated header field) can be run through a hash function and reattached to the header, for example, to enable anonymous identification of messages from the same origin as may be desired.
  • users can be notified of the targeted asset system and allowed to opt in or opt out such that participating users have positively assented to participate.
  • FIG. 7A illustrates a theoretical example of a CPE including a television set 1100 and a DSTB 1102 that are associated with multiple users 1103 - 1106 .
  • Arrow 1107 represents a user input stream, such as a click stream from a remote control, over time.
  • a first user 1105 in this case a child, uses the television 1100 during a first time period—for example, in the morning.
  • Second and third users 1103 and 1104 (designated “father” and “mother”) use the television during time periods 1109 and 1110 , which may be, for example, in the afternoon or evening.
  • a babysitter 1106 uses the television during a nighttime period in this example.
  • targeting based on household demographics would have limited effectiveness. For example, it may be assumed that the child 1105 and father 1103 in many cases would not be targeted by the same asset providers. Moreover, in some cases, multiple users may watch the same television at the same time as indicated by the overlap of time periods 1109 - 1110 . In addition, in some cases such as illustrated by the babysitter 1106 an unexpected user (from the perspective of the targeted asset system) may use the television 1100 .
  • the system should preferably be operative to distinguish between multiple users of a single set and, in the context of the system described above, vote and report to the network accordingly.
  • the system should preferably react over time to changing conditions such as the transitions from use by father 1103 to use by both father and mother 1103 and 1104 to use by only mother 1104 .
  • the system should also preferably have some ability to characterize unexpected users such as the babysitter 1106 . In that case, the system may have no other information to go on other than the click stream 1107 .
  • the system may also identify time periods where, apparently, no user is present, though the set 1100 may still be on.
  • the system also operates free from persistent storage of any user profile or sensitive information so that no third party has a meaningful opportunity to misappropriate such information or discover the private network usage patterns of any of the users 1103 - 1106 via the targeted asset system.
  • Privacy concerns can alternatively be addressed by obtaining consent from users.
  • sensitive information including PII can be transmitted across the network and persistently stored for use in targeting. This may allow for compiling a detailed user profile, e.g., at the headend. Assets can then be selected based on the user profile and, in certain implementations, addressed to specific CPEs.
  • the present invention monitors the click stream over a time window and applies a mathematical model to match a pattern defined by the click stream to predefined audience classification parameters that may relate to demographic or psychographic categories.
  • the click stream will indicate programs selected by users, volume and other information that may have some correlation, at least in a statistical sense, to the classification parameters.
  • factors such as the frequency of channel changes and the length of time that the user lingers on a particular asset may be relevant to determining a value of an audience classification parameter.
  • the system can also identify instances where there is apparently no user present.
  • logic associated with the CPE 1101 uses probabilistic modeling, fuzzy logic and/or machine learning to progressively estimate the audience classification parameter values of a current user or users based on the click stream 1107 .
  • This process may optionally be supplemental based on stored information (preferably free of sensitive information) concerning the household that may, for example, affect probabilities associated with particular inputs.
  • each user input event (which involves one or more items of change of status and/or duration information) can be used to update a current estimate of the audience classification parameters based on associated probability values.
  • the fuzzy logic may involve fuzzy data sets and probabilistic algorithms that accommodate estimations based on inputs of varying and limited predictive value.
  • the click stream is modeled as an incomplete or noisy signal that can be processed to obtain audience classification parameter information. More specifically, a series of clicks over time or associated information can be viewed as a time-based signal.
  • This input signal is assumed to reflect a desired signature or pattern that can be correlated to audience classification parameters. However, the signal is assumed to be incomplete or noisy—a common problem in signal processing. Accordingly, filtering techniques are employed to estimate the “true” signal from the input stream and associated algorithms correlate that signal to the desired audience classification information. For example, a nonlinear adaptive filter may be used in this regard.
  • Modeling user inputs e.g., a clickstream
  • a current audience classification e.g., in terms of classification parameters
  • the inputs into the system are primarily a click stream and stored aggregated or statistical data, substantially free of any sensitive information.
  • This addresses privacy concerns as noted above but also provides substantial flexibility to assess new environments such as unexpected users.
  • the system preferably has a forgetfulness such that recent inputs are more important than older inputs. Either of the noted examples accommodates this objective. It will be appreciated that such forgetfulness allows the system to adapt to change, e.g., from a first user to multiple users to a second user.
  • such forgetfulness limits the amount of viewing information that is available in the system at any one time, thereby further addressing privacy concerns, and limits the time period during which such information could conceivably be discovered. For example, information may be deleted and settings may be reset to default values periodically, for example, when the DSTB is unplugged.
  • FIG. 7B A block diagram of a system implementing such a user classification system is shown in FIG. 7B .
  • the illustrated system is implemented in a CPE 1120 including a user input module 1122 and a classification module 1124 .
  • the user input module receives user inputs, e.g., from a remote control or television control buttons, that may indicate channel selections, volume settings and the like. These inputs are used together with programming information 1132 (which allows for correlation of channel selections to programming and/or associated audience profiles) for a number of functions.
  • programming information 1132 which allows for correlation of channel selections to programming and/or associated audience profiles
  • the presence detector 1126 determines whether it is likely that a user is present for all or a portion of an asset that is delivered.
  • the classifier 1128 develops audience classification parameters for one or more users of a household as discussed above.
  • the user identifier is operative to estimate which user, of the classified users, is currently present.
  • these modules 1126 , 1128 and 1130 provide audience classification information that can be used to vote (or elect not to vote) and/or generate reports (or elect not to generate reports).
  • one of the audience classifications that may be used for targeting is location.
  • an asset provider may wish to target only users within a defined geographic zone (e.g., proximate to a business outlet) or may wish to target different assets to different geographic zones (e.g., targeting different car ads to users having different supposed income levels based on location).
  • the present invention determines the location of a particular CPE and uses the location information to target assets to the particular CPE. It will be appreciated that an indication of the location of a CPE contains information that may be considered sensitive.
  • the present invention also creates, extracts and/or receives the location information in a manner that addresses these privacy concerns. This may also be accomplished by generalizing or otherwise filtering out sensitive information from the location information sent across the network.
  • logic associated with the CPE sends an identifier upstream to the headend where the identifier is cross-referenced against a list of billing addresses.
  • the billing address that matches the identifier is then translated, for example, using GIS information, into a set of coordinates (e.g., Cartesian geographic coordinates) and those coordinates or an associated geographic zone identifier are sent back to the CPE for storage as part of its location information.
  • a list may be broadcast. In this case, a list including location information for multiple or all network users is broadcast and each CPE selects its own information.
  • Asset providers can also associate target location information with an asset. For example, in connection with a contract interface as specified below, asset providers can define target asset delivery zones.
  • this can be done via a graphical interface (e.g., displaying a map), and the defined zones can match, to a fine level of granularity, targeted areas of interest without being limited to node areas or other network topology.
  • a graphical interface e.g., displaying a map
  • the defined zones can match, to a fine level of granularity, targeted areas of interest without being limited to node areas or other network topology.
  • such zones can have complex shapes including discontiguous portions.
  • the zones can then be expressed in terms that allow for convenient transmission in asset metadata and comparison to user locations e.g., in terms of grid elements or area cells.
  • individual geographic regions are associated with unique identifiers and new regions can be defined based on the union of existing regions. This can be extended to a granularity identifying individual CPEs at its most fine level. Higher levels including numerous CPEs may be used for voting and reporting to address privacy concerns.
  • the CPE Upon receipt of an asset option list or an asset delivery request (ADR), the CPE parses the ADR and determines whether the location of the CPE is included in the locations targeted by the asset referenced in the ADR. For example, this may involve a point in polygon or other point in area algorithm, a radius analysis, or a comparison to a network of defined grid or cells such as a quadtree data structure. The CPE may then vote for assets to be received based on criteria including whether the location of that particular CPE is targeted by the asset.
  • the CPE may also use its location information in the reporting process to enhance the delivery data sent upstream.
  • the process by which the CPE uses its location information removes substantially all sensitive information from the location information. For example, the CPE may report that an asset targeted to a particular group of locations was delivered to one of the locations in the group. The CPE in this example would not report the location to which asset was actually delivered.
  • tags are additional information that is superimposed on or appended to such assets.
  • a tag may provide information regarding a local store or other business location at the conclusion of an asset that is distributed on a broader basis.
  • tags have been appended to ads prior to insertion at the headend and have been limited to coarse targeting.
  • tags may be targeted to users in particular zones, locations or areas, such as neighborhoods.
  • tags may also be targeted based on other audience classification parameters such as age, gender, income level, etc.
  • tags at the end of a department store ad may advertise specials on particular items of interest to particular demographics.
  • a tag may be included in an asset flotilla and conditionally inserted based on logic contained within the CPE.
  • the tags are separate units that can be targeted like other assets, however, with conditional logic such that they are associated with the corresponding asset.
  • the present invention may use information relating to the location of a particular CPE to target a tag to a particular CPE.
  • the CPE may contain information relating to its location in the form of Cartesian coordinates as discussed above. If an asset indicates that a tag may be delivered with it or instead of it, the CPE determines whether there is, associated with any of the potential tags, a location criterion that is met by the location information contained in the particular CPE.
  • a tag may include a location criterion defining a particular neighborhood. If the CPE is located in that neighborhood, the CPE 1101 may choose to deliver the tag, assuming that other criteria necessary for the delivery of the tag are met. Other criteria may include the time available in the given break, other demographic information, and information relating to the national or non-localized asset.
  • targeting may also be implemented based on marketing labels.
  • the headend may acquire information or marketing labels regarding a user or household from a variety of sources. These marketing labels may indicate that a user buys expensive cars, is a male, is 18-24 years old, and/or include other information of potential interest to an asset provider. In some cases, this information may be similar to the audience classification parameters, though it may optionally be static (not varying as television users change) and based on hard data (as opposed to being surmised based on viewing patterns or the like). In other cases, the marketing labels may be more specific or otherwise different than the audience classification. In any event, the headend may inform the CPE as to what kind of user/household it is in terms of marketing labels. An asset provider can then target an asset based on the marketing labels and the asset will be delivered by CPEs where targeting matches. This can be used in audience aggregation and spot optimization contexts.
  • the targeted asset system of the present invention allows for targeting of assets in a broadcast network based on any relevant audience classification, whether determined based on user inputs such as a click stream, based on marketing labels or other information pushed to the customer premises equipment, based on demographic or other information stored or processed at the headend, or based on combinations of the above or other information.
  • targeting concepts that have previously been limited to other contexts such as direct mail. For example, such targeting may make use of financial information, previous purchase information, periodical subscription information and the like.
  • classification systems developed in other contexts may be leveraged to enhance the value of targeting achieved in accordance with the present invention.
  • determining classification parameters for a user and matching the classification parameters of the user to targeting parameters of an asset is useful in several contexts.
  • this matching-related functionality is useful in the voting process. That is, one of the functions of the targeting system in the system described above is to receive ad lists (identifying a set of ads that are available for an upcoming spot), determining the targeting parameters for the various ads and voting for one or more ads based on how well the targeting parameters match the classification parameters of a current users.
  • identifying the classification parameters of the current user(s) and matching those parameters to the targeting parameters is important in the voting context.
  • Matching related functionality is also important in the ad selection context. Specifically, after the votes from the various participating set top boxes have been processed, a flotilla of ads is assembled for a commercial break. A given DSTB selects a path through the flotilla (corresponding to a set of ads delivered by the set top box at the commercial break) based on which ads are appropriate for the user(s). Accordingly, identifying the current user classification parameters and matching those parameters to the targeting parameters is important in the ad selection context.
  • the matching related functionality may also be used in the reporting context.
  • some or all at the DSTBs provide reports to the network concerning the ads that were actually delivered. This enables the targeting system and the traffic and billing system to measure the audience for an ad so that the advertiser can be billed appropriately.
  • the information provided by these reports not only indicates the size of the audience but how well the audience fits the target audience for the ad. Accordingly, the system described above can provide goodness of fit information identifying how well the classification parameters of the user(s) who received the ad match the targeting parameters.
  • the matching related functionality is also useful in this context.
  • this matching related functionality is performed by a classifier resident at the DSTB.
  • This classifier will be described in more detail below.
  • the classifier analyzes a click stream, or series of remote control inputs, to determine probable classification parameters of a current user(s).
  • the classifier also performs a matching function to determine a suitability of each of multiple candidate ads (e.g., from an ad list) for the current user(s) based on inferred classification parameters of a putative user or users.
  • the classifier can then provide matching related information for use in the voting, ad selection and reporting contexts as described above. In the description below, this matching related information is primarily discussed in relation to the voting context.
  • the classifier generally operates in two modes: the learning mode and the working mode. It will be appreciated, however, that these modes are not fully separate. For example, the classifier continues to learn and adapt during normal operation. These modes are generally illustrated in FIG. 9 .
  • the illustrated classifier 1300 monitors behavior of viewers 1302 in the audience of a given DSTB 1304 to deduce classification parameters for the viewers 1302 .
  • an audience for a given DSTB 1304 may include a father, a mother and a child, one or more of whom may be present during a viewing session.
  • the classification parameters may include any of the classification parameters noted above, such as gender, age, income, program preferences or the like.
  • the classifier receives inputs including click data 1310 from the user, program data 1312 (such as program guide data) from the network and Nielsen data 1314 generated by the Nielsen system.
  • This information is processed to learn certain behaviors of the viewer, including a viewer program selection behavior.
  • the Nielsen data 1314 reflects the demographic composition for particular programs.
  • the program data 1312 may include information regarding the genre, rating, scheduled time, channel and other information regarding programs.
  • the click data 1310 reflects channels selected by a user, dwell time (how long a user remained on a given channel) and other information that can be correlated with the Nielsen data 1314 and program data 1312 to obtain evidence regarding classification parameters of the viewer.
  • the click data 1310 may reflect frequency of channel hopping, quickness of the click process, volume control selections and other information from which evidence of viewer classification parameters may be inferred independent of any correlation to program related information.
  • the classifier 1300 may begin to generate clusters of data around segments of classification parameter values (e.g., associated with conventional data groupings), thereby learning to identify viewers 1310 and classify the viewers 1302 in relation to their probable gender, age, income and other classification parameters.
  • the state transition functionality is illustrated in FIG. 10 .
  • State changes are triggered by events, messages and transactions.
  • One of the important state transitions is the stream of click events 1400 .
  • Each click event 1400 represents a state transition (for example, a change from one program to another, a change in volume setting, etc.).
  • an absence of click events 1400 or low frequency of click events may indicate that no viewer is present or that any viewing is only passive 1404 .
  • the transition count 1406 exceeds a threshold (e.g., in terms of frequency) and the DSTB is on or active 1402 , then the classifier may be operated to match program data 1408 and learn or store 1410 the program. Programs may be deleted 1412 in this regard so as not to exceed a maximum stack depth or to implement a degree of desirable forgetfulness as described above.
  • viewer identifications are developed in relation to at least two sets of characteristic information.
  • a classification parameter set is developed for each discovered viewer 1302 of the DSTB audience.
  • a set of rules is developed that defines the viewing behavior over time for that viewer 1302 . This is referred to below as the periodicity of the viewer's viewing habits.
  • the classification set for each discovered viewer 1302 may identify the viewer's age, gender, education, income and other classification parameters. This information is coupled with the periodicity of the viewer's viewing habits so as to allow the classifier to match an ad with a target audience during a specific timeframe.
  • the determination by the classifier as to who is watching at a given time may be informed both by a substantially real-time analysis of viewing behavior and by historical viewing patterns of a viewer 1302 .
  • this process of developing classification parameter sets for discovered users may take into consideration multiple time frames, e.g., different times of day. Developing these classification parameter sets for discovered viewers as a function of time of day, e.g., on an hourly, half-hourly or other time dependent basis, has been found effective, as viewership in many households is significantly dependent on time of day.
  • the classifier may estimate classification parameters for a current viewer 1302 even if historical periodicity data has not been developed for that viewer 1302 . Similarly, even where such information has been developed in the learning mode, a current viewing audience may be continually defined and redefined during the working mode.
  • the classifier does not require persistent storage of viewer profile information in order to function. For example, any such stored information may be deleted when the DSTB is turned off. In such a case, the classifier can readily develop classification parameters for one or more viewers when power is restored to the set top box.
  • the classifier may be designed to incorporate a degree of forgetfulness. That is, the classifier may optionally de-weight or delete aged information from its evolving model of audience members. In this manner, the classifier can adapt to changes in the audience composition and identify previously unknown audience members.
  • the illustrated classifier moves from the learning mode to the working mode.
  • the classifier 1306 performs a number of related functions. First, it can receive ad lists 1308 for an upcoming commercial break, match the targeting parameters for the ads on that list 1308 to the classification parameters of the current viewer 1302 , and vote for appropriate ads. The classifier 1306 also selects ads available in a given ad flotilla for delivery to the current user 1302 . Moreover, the classifier 1306 can report goodness of fit information regarding ads delivered during one or more commercial breaks. Again, in the working mode, the classifier 1306 continues to learn through a process of stochastic reinforcement, but the classifier 1306 is deemed to have sufficient information to meaningfully estimate classification parameters of a current viewer 1302 or audience.
  • the classifier 1306 controls the voting process by effectively ranking ads from an ad list 1308 .
  • the illustrated working mode classifier 1306 receives information regarding available ads 1316 from an ad repository 1318 .
  • These ads 1316 are associated with targeting parameters, for example, in the form of audience segmentation and viewer profile classification rules 1320 .
  • targeting parameters for example, in the form of audience segmentation and viewer profile classification rules 1320 .
  • an advertiser may enter targeting parameters directly into the T & B system via the ad interface.
  • these targeting parameters may be defined in relation to conventional audience segmentation categories.
  • the targeting system of the present invention may accommodate different or finer targeting parameters.
  • the working mode classifier 1306 also receives an ad list 1308 or a view list of candidate ads, as described above.
  • the headend targeting system component 1320 processes the inputs regarding available ads and their targeting parameters to generate the ad list 1308 for distribution to participating DSTBs.
  • the similarity and proximity analyzer 1322 uses the targeting parameters associated with individual ads and the classification parameters of the current user to execute matching functionality. That is, the analyzer 1322 matches an ad with at least one of the probable viewers 1302 currently thought to be sitting in front of the television set 1324 . As will be described in more detail below, this is done by comparing, for example, the target age range for an ad (which may be expressed as a slightly fuzzified region) to the set representing the viewer's age (which may also be a fuzzy set). The more these two sets overlap the greater the compatibility or match. Such matching is performed in multiple dimensions relating to multiple targeting/classification parameters.
  • This similarity analysis is applied across each candidate ad of the ad list 1308 , and a degree of similarity is determined for each ad.
  • the ads in each time period can be sorted, e.g., in descending order by similarity, and one or more of the top ads may be selected for voting.
  • the passive voting agent 1326 is operative to select ads based on the match information. This process works in the background generally using an out of band data stream. More specifically, the illustrated voting agent 1326 selects a record or ADR for each of the candidate ads and determines for each if any viewers are likely to be present at the ad time. Additionally, the voting agent 1326 determines if any such viewer has classification parameters that acceptably match the targeting parameters for the ad. In the voting context, for each match, a vote is made for the ad. This vote is returned to the headend component 1320 where it is combined with other votes. These aggregated votes are used to generate the next generation of ad lists 1308 .
  • the learning mode classifier develops classification parameter information for probable viewers as well as periodicity information for those viewers. This process is illustrated in more detail in FIG. 11 .
  • the classifier 1500 is constructing a statistical model of the audience.
  • this process is illustrated with respect to two examples of the classification parameters; namely, target gender and target age.
  • the learning mode classifier 1500 receives inputs including programming viewing frequency (or demographic) data from Nielsen, BBM or another ratings system (or based on previously reported information of the targeted advertising system) 1504 , program data 1506 and click data 1508 . Based on this data, the classifier 1500 can determine which program, if any, is being viewed at a particular time and what that indicates regarding probable classification parameters of a viewer. Additionally, the click data 1508 , since it is an event stream, may indicate the level of focus or concentration of the viewer 1502 at any time and may provide a measure of the level of interest in a particular program. The click data 1508 also allows the classifier 1500 to determine when the DSTB set is turned off and when it is turned on.
  • programming viewing frequency (or demographic) data from Nielsen, BBM or another ratings system (or based on previously reported information of the targeted advertising system) 1504 program data 1506 and click data 1508 . Based on this data, the classifier 1500 can determine which program, if any, is being viewed at a particular time and what that indicates regarding probable
  • the learning mode classifier 1500 fuses the incoming data through a set of clustering and partitioning techniques designed to uncover the underlying patterns in the data.
  • the goal in this regard is to discover the number of probable viewers, build a classification parameter set for each viewer and determine the viewing habits of each viewer over time (or to develop classification parameter sets for likely viewers as a function of time).
  • This information allows the classifier 1500 to determine what kind of audience probably exists at the delivery time for a specific ad.
  • this learning process involves the development of two classifier modules—the age and gender classifier module 1510 and the viewing behavior or periodicity classifier module 1512 .
  • the periodicity classifier 1512 accumulates and reinforces the results of the age and gender classification module across time.
  • a sequence of age interval classes 1510 are stored across an independent axis representing the time of day and day of week that that evidence is collected. This time axis is used to determine the time of day that each individual detected in the age and gender classifier module 1510 tends to watch television.
  • the gender and age classifier module 1510 gathers evidence over time as to probable viewers. Once sufficient evidence is collected, it is expected that the evidence will cluster in ways that indicate a number of separate audience members. This is at once a fairly simple and complex process. It is simple because the core algorithms used to match viewing habits to putative age and gender features are well understood and fairly easy to implement. For example, it is not difficult to associate a program selection with a probability that the viewer falls into certain demographic categories. On the other hand, it is somewhat complex to analyze the interplay among parameters and to handle subtle phenomena associated with the strength or weakness of the incoming signals. In the latter regard, two parameters that affect the learning process are dwell time and Nielsen population size.
  • the dwell time relates to the length of time that a viewer remains on a given program and is used to develop an indication of a level of interest.
  • dwell time functions like a filter on the click stream events that are used in the training mechanism.
  • one or more thresholds may be set with respect to dwell time to attenuate or exclude data.
  • the classifier does not benefit from learning that a viewer watched a program if the viewer watched that program for less than a minute or, perhaps, less than 10 seconds.
  • Thresholds and associated attenuation or exclusion factors may be developed theoretically or empirically in this regard so as to enhance identification accuracy.
  • audience expectation measure This is the degree to which, at any time, it is expected that the television (when the DSTB is turned on) will have an active audience. That is, it is not necessarily desirable to have the classifier learn what program was tuned in if nobody is in fact watching.
  • the audience expectation measure can be determined in a variety of ways. One simple measure of this factor is the number of continuous shows that has elapsed since the last channel change or other click event. That is, as the length of time between click events increases, the confidence that someone is actively watching decreases. This audience expectation measure can be used to exclude or attenuate data as a factor in developing a viewer identification model.
  • the degree of membership in a particular age group is treated as evidence for that age.
  • the gender distribution may contain a broad spectrum of viewer population frequencies.
  • the dwellage percentage of the audience that falls into each age group category is also evidence for that age group. Accordingly, the amount that a set is incremented is scaled by the degree of membership with respect to that set. Thus, for example, if few viewers of an age category are watching a program, this is reflected in only a small amount of evidence that the viewer is in this age group.
  • the illustrated learning mode classifier 1500 also encompasses a periodicity classifier module 1512 .
  • a periodicity classifier module 1512 As the classifier 1500 develops evidence that allows for determining the number of viewers in an audience of a DSTB and for distinguishing between the viewers, it is possible to develop a viewing model with respect to time for each of these viewers. This information can then be used to directly predict who is likely to be watching at the time of ad delivery.
  • this model may involve mapping a viewer to their pattern. Alternatively, a pattern may be discovered and then matched to a known viewer. In the illustrated implementation, the latter viewer-to-pattern approach is utilized. As will be understood from the description below, this approach works well because the properties that define the periodicity are fuzzy numbers. The match can therefore use the same kind of similarity function that is used to match viewers to targeting parameters of ads.
  • viewers are identified based on integrating or aggregating evidence in relation to certain (e.g., fuzzy) sets.
  • This process may be more fully understood by reference to FIG. 12 .
  • this which for purposes of illustration is limited to discovery of age and gender, this involves a 2 ⁇ M fuzzy pattern discovery matrix.
  • the two rows are gender segmentation vectors.
  • the M columns are conventional age intervals used in ad targeting. These age intervals are overlaid with fuzzy interval measures.
  • the classifier is modeled around certain age intervals (12-17, 25-35, 35-49) because these are industry standard segmentations. It will be appreciated, however, that specific age groups are not a required feature of the classifier.
  • the fuzzy intervals are represented by the trapezoidal fuzzy set brackets that illustrate a certain amount of overlap between neighboring age intervals. This overlap may improve discrimination as between different age ranges.
  • the bars shown on the matrix reflect the accumulation of evidence based on a series of click events. As can be seen in the matrix of FIG. 12 , over time, this evidence tends to cluster in a fashion that indicates discrete, identifiable viewers associated with different classification parameters.
  • this process of accumulating evidence to identify discrete viewers of an audience is depicted in three-dimensional graphics.
  • the result of the learning process is a collection of gradients or hills or mountains in the learning matrix or multi-dimensional (in this case, two-dimensional) feature terrain.
  • the higher hills or “mountains” with their higher gradient elevations provide the best evidence that their site in the learning matrix is the site of a viewer.
  • the hills are constructed from the counts in each of the cells defined by the classification parameter segmentation (e.g., age groups). The greater the count, the higher the hill and, consequently, the more certain it is that the characteristics correspond to a viewer of the audience.
  • the gradients and their elevations form a topological map, as shown in FIG. 14 .
  • the concentration and height of the contour lines reflect a clustering that suggests discrete and identifiable viewers.
  • the map of FIG. 14 reflects three probable viewers identified from a learning process.
  • the fuzzy terrain mapping which allows certain degree of overlap between surrounding age groups, provides a more refined estimate of the actual classification parameter values of a viewer. That is, an interpolation of evidence in adjacent fuzzy sets enables an estimation of an actual parameter value that is not limited to the set definitions. This interpolation need not be linear and may, for example, be executed as a center of gravity interpolation. That takes into account gradient height in each fuzzy set as well as a scaled degree of membership of the height in the set.
  • the process for applying evidence to the terrain may thus result in a fuzzified terrain. That is, rather than simply applying a “count” to a cell of the terrain based on determined classification parameter values, a count can effectively be added as terrain feature (e.g., a hill) centered on a cell (or on or near a cell boundary) but including residual values that spill over to adjacent cells. The residual values may spill over onto adjacent cells in multiple dimensions. In one implementation, this effect is defined by a proximity calculation.
  • the result of applying evidence to the terrain in a fuzzified fashion is potentially enhanced user definition as well as enhanced ability to distinguish as between multiple users.
  • the proximity algorithm and terrain seeding noise reduction filtering can be illustrated by reference to an example involving three classification parameter dimensions; namely, gender, age and time of day.
  • the feature terrain may be viewed as being defined by a cube composed of a number of sub-cubes or grains.
  • the terrain cube may be composed of 7680 grains—2 ⁇ 80 ⁇ 48 grains, corresponding to two gender, 80 age and 48 half-hour time categories.
  • Each of the grains can be populated with a reference or noise value.
  • the noise values can be derived from a statistical analysis of the expected viewers at each age, gender and time measured over all available programming channels for which Nielson, BBM or another ratings system have observers (or based on previously reported information of the targeted advertising system).
  • the noise may be drawn from a distribution of these viewers from a suitable function (e.g., developed empirically or theoretically) and may or may not be as simple as an average or weighted average. In any case, it is this noise bias that cancels out a corresponding randomness in the evidence leaving a trace or residue of evidence only when the evidence is repeatedly associated with an actual viewer whose behavior matches the (age, gender, time) coordinates of the terrain.
  • a suitable function e.g., developed empirically or theoretically
  • the grain definition and population of the grains with reference values or noise is not limited to the granularity of the source of the reference values, e.g., the standard Neilson categories. Rather, the reference values can be interpolated or estimated to match the defined grain size of the terrain cube.
  • Neilson source data may be provided in relation to 16 age gender groups whereas the terrain cube, as noted above, may include 2 ⁇ 80 corresponding grains (or columns of grains where the column axis corresponds to the time dimension).
  • the age distribution can be fitted to a curve or function, which can then be solved for each age value. The corresponding values are then applied to seed the terrain.
  • Evidence is the statistical frequency of viewers of a particular age and gender who are watching a television program (which is playing at a particular time).
  • the combination of Neilson frequencies and the program time generates 16 pieces of evidence for each time period (there are 8 male and 8 female age groups with their viewing frequencies). Finding out which of these 16 pieces of evidence corresponds to the actual viewer is the job of the statistical learning model underlying the classifier. Fundamental to making this decision is the methodology used to add evidence to the terrain. This involves first applying the noted noise filter as follows:
  • the number of positive and negative residuals (d) will be approximately equal, and the total sum of the terrain value (t) at that point will also be zero. If, however, the evidence is associated with an actual viewer, then we would expect a small but persistent bias in the frequency statistics to accumulate around that real viewer. Over time, this means that residual (d) will be positive more often than negative and that the contour at the terrain cube will begin to grow. As more evidence is added, the contour grows as a small hill on the terrain. This small hill is an actual viewer (actually, because of the time axis, viewers appear as a ridge of connected hills, somewhat like a winding mountain range).
  • the value at each outward layer is based on a fraction of the previous layer (not the original value).
  • the original value is very small, so the values in each outward cell become very, very small very, very quickly. Yet, over time, they provide enough additional evidence to support the growth of a valid contour site.
  • the proximity algorithm that fills out the terrain also provides a quick and effective way of discovering who is viewing the television associated with the DSTB.
  • the ridge of contours or hills associated with a viewer wanders across the three-dimensional terrain cube. These hills are smoothed out (made “fatter,” so to speak) over the terrain by the addition of the minute bits of partial evidence laid down during the building of the terrain (the x1, x2, x3, etc., in the previous section).
  • the hills over the gender/age axes at a particular time can be summed and averaged. The viewer is the hill with the maximum average height.
  • a proximity algorithm can be used in integrating these hills.
  • a fraction of the height of adjacent grains and a small fraction of the next outward layer of grains may be added. This is somewhat analogous to the mountain clustering algorithm discussed herein and may be used as an alternative thereto.
  • these terrains may be developed as a function of time of day.
  • the data may be deposited in “bins” that collect data from different times of day on an hourly, half-hourly or other basis (e.g., irregular intervals matching morning, daytime, evening news, primetime, late night, etc.).
  • the process of applying the evidence to these bins may be fuzzified such that evidence spills over to some extent into adjacent bins or cells in the time and other dimensions.
  • Evidence may be integrated in these bins over multiple days.
  • the resulting terrains may be conceptualized as multiple terrains corresponding to the separate timeframes or as a single terrain with a time dimension. This functionality may be implemented as an alternative to the separate periodicity analysis described below.
  • FIG. 15 illustrates a more difficult dataset in this regard.
  • the feature terrain of FIG. 15 does not readily yield interpretation as to the number of viewers in the audience or their specific classification parameters. Any number of factors may result in such complexity.
  • the remote control may be passed between different audience members, the click stream may be influenced by other audience members, a click event of significant duration may reflect distraction rather than interest, a viewer may have a range of programming interests that do not cleanly reflect their classification parameters, etc.
  • the classifier implements processes of noise removal and renormalization of the gradient terrain.
  • the goal is to find the actual centers of evidence so that the number of viewers and their classification parameters can be accurately identified.
  • One type of process that may be implemented for removing or attenuating noise involves consideration of reference values, e.g., average values for all events at that time, taken in relation to the whole audience.
  • the terrain may be seeded with reference values or the reference values may be considered in qualifying or rejecting data corresponding to individual events.
  • data is compared to reference values on an event-by-event basis to qualify data for deposit into the bins for use in developing the terrain. This has the effect of rejecting data deemed likely to represent noise.
  • the reference values are subtracted from evidence as it is applied to the terrain, thus impeding the process of constructing terrain features so that terrain features substantially only rise as a result of persistent or coherent accumulation of evidence likely reflective of an actual user, and spurious peaks are avoided.
  • Selecting the reference values as average values, weighted average values or some other values related to the observation context (but substantially without information likely to bias the user identification and definition problem) has the effect of scaling the filter effect to properly address noise without inhibiting meaningful terrain construction.
  • data may be qualified in relation to a presence detector. As noted above, presence may be indicated by reference to the on/off state of the DSTB and/or by the absence or infrequency of inputs over a period of time. Data acquired when the presence detector indicates that no user is deemed “present” may be excluded.
  • Noise removal may further involve eliminating false centers and sporadic evidence counts without disturbing the actual centers to the extent possible.
  • FIG. 16 illustrates a possible identification of viewer sites with respect to the data of FIG. 15 .
  • FIG. 17 illustrates an alternative, also potentially valid, interpretation of the data of FIG. 15 .
  • the classifier implements an algorithm designed to determine which of competing potentially valid interpretations is most likely correct.
  • This algorithm generally involves gradient deconstruction.
  • the deconstruction process is an iterative process that finds gradient centers by first removing low-level interference noise (thus revealing the candidate hills) and then measuring the compactness of the distribution of hills around the site area.
  • a mountain clustering algorithm can be implemented to iteratively identify peaks, remove peaks and revise the terrain. The affect is to identify cluster centers and to smooth out the hilliness between cluster centers so that the cluster centers become increasingly distinct.
  • the remaining sites after the feature terrain is processed in this regard are the sites of the putative viewers.
  • FIG. 18 shows a learning matrix or feature terrain with a set of gradients scattered over the surface.
  • gradient C 1 is identified as the maximum mountain.
  • the distance from C 1 's center of gravity to the centers of each gradient on the surface (e.g., C 2 , C 3 , etc.) is measured.
  • the degree of mountain deconstruction on each gradient is inversely proportional to the square distance between C 1 and each of the other clusters. This inverse square mechanism localizes nearly all of the mountain deconstruction to the neighborhood of C 1 .
  • FIG. 20 shows significant deconstruction of features proximate to C 1 as a result of this first iteration. This causes the general perimeter of the emerging site at C 1 to contract. The contraction also involves the assimilation of small gradients, hence the height of C 1 is increased, and the adjacent hilliness is reduced. At the same time, the height of C 3 is barely changed and its set of satellite gradients has not yet begun to be assimilated. The process is then repeated with respect to C 3 . After a number of iterations, the final sites begin to emerge and stabilize.
  • FIG. 21 shows the terrain after most of the smaller gradients have been assimilated. The resulting processed terrain or feature space has well defined gradients at each putative viewer site. The mountain clustering process thus essentially removes much ambiguity.
  • a similar process is performed with respect to the periodicity analysis. Specifically, a periodicity learning matrix is created and updated in a manner analogous to construction and updating of the viewer classification parameter matrix and its conversion to the feature terrain.
  • the periodicity terrain produces a set of gradients defining, by their height and width, the expectation that the viewer is watching television at that point in time.
  • a matching algorithm can then be used to match a periodicity pattern to one of identified viewers.
  • the classifier When the classifier has been sufficiently trained, it moves from learning mode operation to working mode operation. As noted above, these modes are not entirely distinct. For example, the classifier can perform estimations of classification parameters while still in the learning mode, and the classifier continues to learn during the working mode. However, as described above, the learning mode is a gradual process of collecting evidence and measuring the degree to which the viewer sites are discernable in the learning matrix. Thus while the classifier can operate quickly using default values, working mode operation reflects a determination that the viewers of an audience have been identified and classified with a high degree of confidence.
  • FIG. 22 The basic operation of the working mode classifier is illustrated in FIG. 22 .
  • the classifier 2600 has access to the developed feature terrain, as discussed above, as well as to a similarity function that is operative to match targeting parameters to the classification parameters as indicated by the viewer identification. More specifically, Nielsen data 2602 and program data 2604 continues to be fed into the classifier 2600 in the working mode to support its continued learning process that essentially runs in the background.
  • the classifier 2600 then receives an ADR from the ad list 2608 .
  • the ADR is initially filtered for high level suitability and then, if the ADS is still available for matching analysis, selects each viewer and each classification parameter for each viewer for comparison.
  • FIG. 22 depicts the process of accessing a classification matrix 2606 of a viewer.
  • the matrix 2606 is a two-dimensional matrix limited to gender and age classification parameters. In practice, any number of classification parameters may be supported.
  • the retrieved classification parameter values of the user and the targeting parameters of the ad are then provided to the similarity and proximity analyzer 2612 where they are compared. This function returns a degree of similarity for each attribute.
  • the total compatibility or relative compatibility rank (RCR) is given as:
  • an ad may be found to be compatible with one or more viewers of an audience. For example, such compatibility may be determined in relation to an RCR threshold value.
  • the periodicity analyzer 2616 is called to see if the viewer is likely to be present at the target ad insertion time. If the viewer is unlikely to be watching, the degree of this time constraint is used to adjust the RCR. Accordingly, the RCR is recomputed as:
  • the RCR will have a high value if there is both a high degree of match between the classification parameters of a user and the targeting parameters of the ad, and there is a high probability that the user will be watching at the ad delivery time.
  • This formula further has the desirable quality that a low compatibility where the viewer is not watching, even if the viewer's classification parameters are a very good match for a given ad.
  • the voting agent 2614 is closely connected to the operation of the working mode classifier 2600 .
  • the validity of the vote is highly dependent on the ability to correctly identify a viewer's classification parameters and viewing habits.
  • the voting agent 2614 essentially works in the same way as the view list ranking Specifically, an ADR is sent to the voting agent 2614 , the voting agent 2614 extracts the ads targeting parameters and calls the classifier 2600 , which returns the RCR for this ad. In this manner, ads are not only voted on but also delivered based on a matching process.
  • the similarity function used to execute the various matching functionality as discussed above can be understood by reference to FIGS. 23-26 .
  • the targeting parameters and any ad constraints are embedded in the ADR.
  • the ADR may also indicate an “importance” of the ad. For example, such importance may be based on the ad pricing (e.g., CPM value) or another factor (e.g., a network operator may specify a high importance for internal marketing, at least for a specified geographic area such as where a competing network is, or is becoming, available).
  • the classifier also has access to the classification parameters of the various viewers in the audience, as discussed above.
  • the first step in the matching process is applying a similarity function, e.g., a fuzzy similarity function, to each of the classification parameters and targeting parameters.
  • the similarity function determines the degree to which a targeting parameter is similar to or compatible with a classification parameter.
  • the weighted average of the aggregated similarity of values for each of the classification/targeting parameters is the base match score.
  • the illustrated classifier fuzzifies the targeting parameter and finds the degree of membership of the corresponding classification parameter of the user in this fuzzified region.
  • the targeting parameters for an ad may specify a target age range of 24-42 years of age.
  • this targeting parameter is redefined as a fuzzy set.
  • the fuzzy set has a small but real membership value across the entire domain of the classification parameter (age in this case).
  • the membership function means that the matching process cannot automatically identify and rank classification parameters values that lie near but, perhaps, not inside the target range.
  • FIG. 24 shows a putative viewer with an inferred age of 26. In this case, the viewer sits well inside the target age range.
  • the degree of membership is therefore 1.0, indicating a complete compatibility with the target age. In fact, any age that is within the target age range will return a matching membership of 1.0. However, the matching process can also deal with situations where retrieved classification parameters do not match the targeting parameter range.
  • FIG. 25 illustrates the case where a viewer is estimated to be 20 years old and is therefore outside of the rigid target range boundaries. If the classifier had Boolean selection rules, this viewer would not be selected.
  • the nature of the fuzzy target space means that viewer 2 is assigned a similarity or compatibility value greater than zero, in this case 0.53.
  • the classifier now has the option of including viewer 2 , knowing that the viewer is moderately compatible with the target range.
  • FIG. 26 shows a slightly different situation—that of a viewer with an inferred age outside of the required age interval.
  • the membership function As illustrated by the membership function, as the viewer age moves away from the identified targeting age interval, the membership function drops off quickly. In this case, viewer 3 has a compatibility of only 0.18, indicating that it would normally be a poor candidate for an ad with a 24-42 age requirement. It is noted, however, that the fuzzy compatibility mechanism means that the classifier can find and rank a viewer if any viewer exists.
  • audience classification parameters (e.g., of a household, of a current audience, etc.) can also be determined or estimated by way of modeling a click stream as an incomplete or noisy signal and processing the signal to obtain such audience classification information.
  • the information obtained via filtering of the noisy signal can be used in a variety of contexts such as for voting, ad selection and/or goodness of fit determinations as described above.
  • the description below describes a filter theory based Head End ad selection system that is an alternative to the noted voting processes.
  • click stream information can be provided to the Head End, or another network platform, where the audience classification parameters may be calculated.
  • the audience classification parameter, ad selection and other functionality can be varied and may be distributed in various ways between the DSTBs, Head End or other platforms.
  • Nonlinear filtering deals with the optimal estimation of the past, present and/or future state of some nonlinear random dynamic process (typically called ‘the signal’) in real-time based on corrupted, distorted or partial data observations of the signal.
  • the signal X t is regarded as a Markov process defined on some probability space ( ⁇ , I,P) and is the solution to some Martingale problem.
  • the filter can provide optimal estimates for not only the current states of the signal but for previous and future states, as well as path segments of the signal:
  • an effective optimal recursive formula is available.
  • This formula is known as the Kalman filter. While the Kalman filter is very efficient in performing its estimates, its use in applications is inherently limited due to the strict description of the signal and observation processes. In the case where the dynamics of the signal are nonlinear, or the observations have non-additive and/or correlated noise, the Kalman filter provides sub-optimal estimates. As a result, other methods are sought out to provide optimal estimates in these more common scenarios.
  • the particle-filtering estimate yields the optimal nonlinear filter estimate.
  • a control parameter ⁇ is introduced to appropriately moderate the amount of resampling performed.
  • this value can be dynamic over time in order to adapt to the current state of the filter as well as the particular application.
  • This filing also included efficient systems to store and compute the quantities required in this algorithm on a computer.
  • a discrete space and amplitude approximation can be used.
  • a discrete space filter is described in detail in U.S. Pat. No. 7,188,048, entitled “Refining Stochastic Grid Filter” (REST Filter), which is incorporated herein by reference.
  • the state space D is partitioned into discrete cells ⁇ c for c in some finite index set C.
  • this space D could be a d-dimensional Euclidean space or some counting measure space.
  • Each cell yields a discretized amplitude known as a “particle count” (denoted as n ⁇ c ), which is used to form the conditional distribution of the discrete space filter:
  • the invention utilized a dynamic interleaved binary index tree to organize the cells with data structures in order to efficiently recursively compute the filter's conditional estimate based on the real-time processing of observations. While this structure was amenable to certain applications, in scenarios where the dimensional complexity of the state space is small, the data structure's overhead can reduce the method's utility.
  • FIG. 28 depicts the overall targeted advertising system.
  • the system is composed of a Head End 4100 and one or more DSTBs 4200 .
  • the DSTBs 4200 are attempting to estimate the conditional probability of the state of potential viewers in household 4205 , including the current member(s) of the household watching television, using the DSTB filter 4202 .
  • the DSTB filter 4202 uses a pair of models 4201 describing the signal (household) and the observations (the click stream data 4206 ).
  • the DSTB filter 4202 is initialized via the setting 4302 downloaded from the Head End 4100 .
  • To estimate the state of the household the DSTB filter 4202 also uses program information 4207 (which may be current, or in the recent past or future), which is available from a store of program information 4208 .
  • program information 4207 which may be current, or in the recent past or future
  • the DSTB filter 4202 passes its conditional distribution or estimates derived thereof to a commercial selection algorithm 4203 , which then determines which commercials 4204 to display to the current viewers based on the filter's output, the downloaded commercials 4301 , and any rules 4302 that govern what commercials are permissible given the viewer estimates.
  • the commercials displayed to the viewers are recorded and stored.
  • the DSTB filter 4202 estimates, as well as commercial delivery statistics and other information, may be randomly sampled 4303 and aggregated 4304 to provide information to the Head End 4100 .
  • This information is used by a Head End filter 4102 , which computes (subject to its available resources) the conditional distribution for the aggregate potential and actual viewership for the set of DSTBs with which it is associated.
  • the Head End filter 4102 uses an aggregate household and DSTB feedback model 4101 to provide its estimates. These estimates are used by the Head End commercial selection system 4103 to determine which commercials should be passed to the set of DSTBs controlled by the Head End 4100 .
  • the commercial selection system 4103 also takes into account any market information 4105 available concerning the current commercial contracts and economics of those contracts.
  • the resulting commercials selected 4301 are subsequently downloaded to the DSTBs 4200 .
  • the commercials selected for downloading affect the level settings 4104 , which provide constraints on certain commercials being shown to certain types of individuals.
  • the signal of a household is modeled as a collection of individuals and a household regime. In one preferred embodiment, this household represents the people who could potentially watch a particular television that uses a DSTB.
  • Each individual (denoted as X i ) at a given point in time t has a state from the state space s ⁇ S, where S represents the set of characteristics that one wishes to determine for each person within a household. For example, in one embodiment one may wish to classify the age, gender, income, and watching status of each individual.
  • certain behavioral information in particular, the amount of television watched by each individual, is useful in developing and using classifications. Age and income may be considered as real values, or as a discrete range.
  • the state space would be defined as:
  • the household regime represents a current viewing “mindset” of the household that can materially influence the generation of click stream data.
  • the household's current regime r t is a value from the state space R.
  • the regimes can consist of values such as “normal,” “channel flipping,” “status checking,” and “favorite surfing.”
  • the rate functions for an individual i depend only on the given individual, the empirical measure of the signal, the current time, and some external environmental variables ⁇ (t, ⁇ t i , ⁇ t ⁇ t ).
  • the number of individuals within the household n t varies over time via birth and death rates. birth and death rates do not merely indicate new beings being born or existing beings dying—they can represent events that cause one or more individuals to enter and exit the household. These rates are calculated based on the current state of all individuals within the household. For example, in one embodiment of the invention a rate function describing the likelihood of a bachelor to have either a roommate or spouse enter the household may be calculated.
  • these rate functions can be formulated as mathematical equations with parameters empirically determined by matching the estimated probability and expected value of state changes from available demographic, macroeconomic, and viewing behavior data.
  • age can be evolved deterministically in a continuous state space such as [0, 120].
  • the observation model describes the random evolution of the click stream information that is generated by one or more individuals' interaction with a DSTB.
  • only current and past channel change information is represented in the observation model.
  • Given a universe of M channels, we have a channel change queue at time t k of Y k (y k , . . . , y k ⁇ B+1 ) with B representing the number of retained channel changes, channels that were watched in the past B discrete time steps.
  • B representing the number of retained channel changes
  • a viewing queue contains this current and past channels as well as such things as volume history.
  • the viewing queue degenerates to the channel change queue.
  • this downloadable content contains, among other things, some program information detailing a qualitative category description of the shows that are currently available, for instance, for each show, whether the show is an “Action Movie” or a “Sitcom”, as well as the duration of the show, the start time of the show, the channel the show is being played on, etc.
  • ⁇ i is the i th random outcome of drawing an element from ⁇ .
  • the observation probabilities that is, the probabilities of switching between two viewing queues over the next discrete step, can be first calculated by determining the probability of switching categories of the programs and then finding the probability of switching into a particular channel within that category.
  • the first step is to calculate, often in a offline manner, the relative proportion of category changes that occur due to channel changes and/or changes in programs on the same channel.
  • C ⁇ c 1 , c 2 , . . . , c K ⁇ .
  • the probabilities for the category transition from c i to c j that occurs at a given time step are calculated first by calculating the probability of category changes given the currently available programs:
  • n t (c j ) is the number of channels that have shows that fall in category c 1 at the end of the current time step.
  • An alternative probability measure may be calculated by the “popularity” of channels instead of the transition between channels at each discrete time step. This above method can be used to provide this form by simply summing over the transition probabilities for a given category:
  • n t (c j ) is the number of channels that have shows that fall into category c j at the end of the current time step.
  • the categories will be programs themselves, given the finest level of granularity. In other instances, it is preferable to have broad categories to reduce the number of probabilities that need to be stored down.
  • D t and X t are the current states of the pertinent exogenous information and signal states at the time of the possible state change.
  • ⁇ ⁇ ⁇ Y 1 , ... ⁇ , Y j ⁇ ] E _ ⁇ ⁇ f ⁇ ( X t ) ⁇ ( Z ⁇ ( T ) ) - 1
  • ⁇ ⁇ ⁇ Y 1 , ... ⁇ , Y j ⁇ ] , ⁇ ⁇ for ⁇ ⁇ t j ⁇ T , where ⁇ ⁇ f : E ⁇ R ⁇ ⁇ and ⁇ ⁇ ⁇ P _ ⁇ ( A ) E ⁇ [ 1 A ⁇ Z ⁇ ( T ) ] ⁇ ⁇ A ⁇ ⁇ ⁇ ⁇ ( X t , Y t ) , t ⁇ T ⁇ .
  • ⁇ _ k ⁇ ( x ) 1 - 1 ⁇ k ⁇ ( x )
  • n s max ⁇ k:t k ⁇ s ⁇ .
  • N is some discretized version of .
  • the application of REST then creates particle counts ⁇ N t c,p ⁇ for each cell in C N and for each household population p within the cell-dependent set of allowable populations P c N , such that
  • ⁇ t N ⁇ ( dx ) ⁇ c ⁇ C N ⁇ ⁇ ⁇ p ⁇ P c N ⁇ ⁇ n t c , p ⁇ ⁇ p , c ⁇ ( dx ) .
  • the signal is composed of zero or more targets X t i and zero or more regimes R t j .
  • each target and regime have only a discrete and finite number of states, and there are a finite number of targets and regimes (and consequently a finite number of possible combinations of targets and regimes). The finite number of combinations need not be all possible combinations—only a finite number of legitimate combinations are required.
  • a finite number of possible types of households can be derived from geography-dependent census information at relatively granular levels. Instead of having all potential combinations of individuals (up to some maximum household membership n MAX ), only those combinations which can be possibly found within a given geographic region need to be considered legitimate and contained within the state space.
  • some components of the state of the target(s) and/or regime(s) may be invariant over the short period during which the optimal estimation is occurring.
  • state information is held to be constant, while other portions of the state information remain variant.
  • the age, gender, income, and education levels of each individual within the household may be considered to be constant, as these values change over longer periods of time and the DSTB estimation occurs over a period of a few weeks.
  • the current watching status and household regime information will change over relatively short time frames, and as a result these states are left to vary in the estimation problem.
  • invariant portion of the signal ⁇ circumflex over (X) ⁇ and the variant portion of the signal as ⁇ tilde over (X) ⁇ .
  • N possible invariant states the i th such state donated by ⁇ circumflex over (X) ⁇ i
  • M i possible variant states for the i th invariant state the j th state denoted by ⁇ circumflex over (X) ⁇ i,j ).
  • FIG. 29 depicts one preferred embodiment of the REST filter in a finite state space environment.
  • REST is composed of a collection of invariant state cells, each of which represents one possible collection of targets and regimes for the signal along with their invariant state properties.
  • Each invariant cell contains a collection of variant state cells, each representing the possible time-variant states of the given invariant cell.
  • the variant cells contain the invariant state information of their parent invariant cell, meaning each variant cell represents a particular potential state of the signal.
  • the invariant cells themselves represent an aggregate container object only and are used for convenience purposes.
  • the collections of variant and invariant cells may be stored on a computer medium in the form of arrays, vectors, list or queues. Cells which have no particle count at a given time t may be removed from such containers to reduce space and computational requirements, although a mechanism to reinsert such cells at a later date is then necessary.
  • each variant state sell contains a particle count n t i,j .
  • This particle count represents the discretized amplitude of that cell. As noted previously, this amplitude is used to calculate the conditional probability of a given state.
  • Each variant state cell also contains a set of imaginary clocks ⁇ t i,j,q . These imaginary clocks represent the time varying progression towards the event of a particle count change within a cell driven by both continuous transition rates and discrete observation events. For each variant state cell there are Q i,j possible state transitions. In this environment, all valid state transitions occur within the same invariant state cell.
  • a temporary particle counter entitled particle count ⁇ n t i,j is used to store the number of particles that will be added or removed from the given variant state cell once the sequential processing of all cells is completed. Cells which have a valid state transition from the variant state cell with state ⁇ tilde over (X) ⁇ i,j are said to be neighbors of that cell.
  • the invariant state cells are containers used to simplify the processing of information.
  • Each invariant state cell's particle count n t i is an aggregate of its child variant state cell particle counts.
  • the invariant state cell's imaginary time clock is an aggregation of all clocks from the variant cells. This aggregation facilitates the filter's evolution, as invariant states which have no current particle count can be skipped at various stages of processing.
  • FIG. 31 depicts the typical evolution of the REST filter.
  • This evolution method updates the conditional distribution of the filter over some time period ⁇ t by transferring particles between neighboring cells using the imaginary clock values.
  • the movement of a particle between neighboring cells is known as an event.
  • Such events are simulated en masse to reduce the computational overhead of the evolution.
  • the number of events to simulate is based on the total imaginary clock sum ⁇ t for all cells.
  • FIG. 32 shows the method that determines how particles move to each neighboring cell. When the simulation of events is complete, the particle counts are updated and the imaginary clocks are scaled back to represent the change in the state of the filter.
  • This method uses some function ⁇ ( ⁇ tilde over (X) ⁇ i,j ,t) to add n t seed particles to variant state cells based on the initial distribution v of the signal. The number of particles to add to each cell depends on time, the given cell, and the overall state of the filter. This method ensures that the filter does not converge to a small set of incorrect states without the ability to recover from an incorrect localization.
  • the determination of which commercials to distribute to a collection of DSTBs is critical. As more information is available about the actual viewership of commercials based on the conditional distributions (or conditional estimates derived thereof) of a DSTB-based asymptotically optimal nonlinear filter, the pricing of specific commercial slots can be more dynamic, thus improving overall profits.
  • an estimate of the collection of household probability distributions that includes such things as the number of people within each demographic, is performed at the Head End based on the whole set or a random sampling of conditional DSTB estimates.
  • the following model contains a prefer embodiment of the Head End estimation system.
  • the Head End signal model consists of pertinent trait information of potential and current television viewers that have DSTB, in communication with a particular Head End.
  • C n ⁇ (( s 1 n 1 ), . . . , ( s r ,n r )): s i ⁇ S and distinct, n 1 +n 2 + . . . +n r +n ⁇ .
  • ⁇ n 0 ⁇ ⁇ C n
  • N is some large number.
  • each DSTB state including potential household viewership, watching status, and current channel; is taken from
  • X we are not worried about a single DSTB nor even which DSTBs are in a particular state but rather with how many DSTBs are in state d ⁇ D. Therefore, we let X to be tracked, be a finite counting measure valued process, counting the number of DSTBs in each category d ⁇ D over time. For technical reasons we define the signal to be either the probability distribution of X of the probability distributions of each component of X.
  • t ⁇ M t ( ⁇ ) is a martingale for each continuous, bounded functional ⁇ on M (D) and is some operator that would be determined largely from the DSTB rates and the natural assumption that the households act independently.
  • V k denote the random selection at time t k in the sampling process.
  • V k would be a matrix with a random number of rows, each row consisting of M entries with exactly one nonzero entry corresponding to the index of the particular DSTB which has provided a sample.
  • the number rows would be the number of DSTBs providing a sample.
  • the locations of the nonzero entries are naturally distinct over the rows and would be chosen uniformly over the possible permutations to reflect the actual sampling taken.
  • ⁇ t k 1 h ( V k ⁇ ( ⁇ circumflex over (P) ⁇ k , U k )).
  • V k would do the random selection and the h would be a function providing the information that is chosen to be communicated to the Head End.
  • the second observation information from the aggregated delivery statistics would be
  • ⁇ t k 2,j H k,j ( ⁇ circumflex over (P) ⁇ t k ⁇ t j ,W k,j ).
  • j ranges back over the spot segments in the reporting periods and t k is the reporting period time.
  • the signal for the Head End is taken to be a representation for the probability distributions from the DSTBs. This assignment can make the estimation problem more workable.
  • aggregate (and possibly delayed) ad delivery statistics can also provide inferences in the estimated viewership of DSTBs, as well as any ‘exposed mode’ information whereby households opt to provide their state information (demographics, psychographics, etc.) in exchange for some compensation.
  • the random arrival of the contract graphs is denoted as the contract graph process.
  • an allotment of resources (that need not be the maximum allotable to any contract) to a contract graph process is called a feasible selection if, given the state (present and future) and the environment, the allotted resources do not exceed the available resources, i.e. the available commercial spots over the various categories.
  • current versus future potential profits are modeled through a utility function.
  • This utility function takes the stream of contract graphs available (both presently and with future random arrivals) and returns a number indicating profit in terms of dollars or some other form of satisfaction. Due to the random future behavior of contract graphs, the utility function cannot simply provide maximum profits without taking into account deviation from the expected profit to ensure the maximization does not allow significant risk of poor profit.
  • the following models need to be defined: the Head End signal model, the Head End observation model, the contract generation model, and the utility (profit) model.
  • the commercial contracts that arise are modeled as a marked point process over the contract graphs.
  • the rate of arrival for the contracts depends upon the previous contracts executed as well as external factors such as economic conditions.
  • n t ( A ) ⁇ 0 ( A )+ ⁇ A ⁇ [0, ⁇ ) ⁇ [0,t] 1 [0, ⁇ (c, ⁇ o,s ,s ))( v ) ⁇ ( dc ⁇ dv ⁇ ds ) for all A ⁇ B ( C ).
  • R(D S ) be the available resources, now and in the future, based upon the downloadable program information D S at time s.
  • n t represents the number of contracts that have arrived of the various types up to and including time t and take
  • ⁇ t ⁇ ( l ) ⁇ Q ⁇ ⁇ C ⁇ [ 0 , t ] ⁇ c ⁇ ( l s - , X s - , q ) ⁇ ⁇ ( ⁇ ⁇ c ⁇ ⁇ ⁇ s ) ⁇ ⁇ q ⁇ ⁇ for ⁇ ⁇ each ⁇ ⁇ t ⁇ 0 ,
  • ⁇ s ⁇ 0 ⁇ is an admissible selection if l s ⁇ R(D s ) for each s ⁇ 0 and l s does not use future contract or observation information, i.e., is measurable with respect to ⁇ ( ⁇ u ,u ⁇ s ⁇ , ⁇ t k 1 , ⁇ t k 2,j j ⁇ N, t k ⁇ s ⁇ ) for each s ⁇ 0.
  • ⁇ t (l) represents the profit obtained up to time t through admissible selection l.
  • be the set of all such admissible selections.
  • the utility function J balances current profit with future profit and the chance of obtaining very high profits on a particular contract with the risk of no or low profit. In order to ensure that we start off reasonably, we will deweight future profit in an exponential manner. Moreover, in order that we are not overly aggressive we will include a variance-like condition.
  • One embodiment of the resulting utility function is
  • the goal of the commercial selection process is to maximize E[J(X,l)] over the l ⁇ ⁇ .
  • Such a goal can be solved using one or more asymptotically optimal filters.
  • a matching algorithm may involve the following steps. First, an ADR is received from matching to one or more of the identified viewers. The similarity function is applied with respect to each parameter, and the results are aggregated as discussed above in order to determine a degree of match. Once the classifier has discovered one or more viewers that match the basic targeting parameters of the ad, the classifier determines whether or not any of these viewers are currently watching the television set. To do this, the periodicity analysis function is called to match a week by day and week by time of day terrain surface to the required time. This process returns a value reflecting the expected degree of match between the viewing time behavior of each viewer and the time frame of the ad. The degree of match in this regard is used to scale the base compatibility score to produce a time compatibility score.
  • the classifier determines whether or not to accept this ad (vote for the ad or select the ad for delivery) based on any frequency limitation constraints and also subject to consideration of ad importance. That is, there may be additional constraints associated with the ad regarding the frequency with which the ad may be delivered to an individual viewer or the total number of times that the ad may be delivered to a viewer.
  • the frequency analyzer returns a value and range of zero to one that is used to scale the time compatibility rank. This creates a frequency compatible rank.
  • a first ad may be selected (and voted) rather than a second ad, where the ads have a similar degree of “match” or even where the second ad has a better match, due to differences in ad importance (e.g., where the first ad has a higher importance).
  • An ad may also have constraints.
  • the ad may have target age limitations, genre restrictions, network restrictions, program nature restrictions, rating restrictions or the like.
  • the placement constraint analyzer that meets the above-noted compatibility requirements and then searches for any required placement constraints.
  • the placement analysis returns a gateway value of one or zero where one indicates no active placement constraints and zero indicates placement constraints that are violated. This creates the final compatibility index score.
  • the matching process need not be based on a continuous, or even finely graded, value range.
  • the result of the matching process may be a binary “match” or “no match” determination.
  • a threshold or set of thresholds with associated decision logic, may be used to define a match or lack thereof for an ad with respect to a current audience.
  • the classifier may thus be viewed as incorporating a number of functional components including a stay transition and link manager, a feature acquisition subsystem, a mountain clustering subsystem, periodicity analysis subsystem and an ad matching and ranking subsystem.
  • the stay transition and link manager are operative to monitor the click stream and determine whether the DSTB is turned on as well as tracking stay changes.
  • the feature acquisition subsystem is operative to build the initial feature terrain as discussed above.
  • the mountain clustering subsystem is operative to process the feature terrain to remove noise and better define viewer sites as discussed above.
  • the periodicity analysis subsystem recognizes viewing patterns and matches those patterns to viewers as identified from the processed feature terrain space.
  • the ad matching and ranking subsystem compares viewer classification parameters to targeting parameters of an ad and also analyzes viewing habits of the viewer in relation to the delivery time, so as to match ads to viewers and develop a ranking system for voting, ad selection and reporting.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Signal Processing (AREA)
  • Multimedia (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Social Psychology (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Marketing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Graphics (AREA)
  • Computing Systems (AREA)
  • Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)

Abstract

A targeted advertising system selects an asset (e.g., ad) for a current user of a user equipment device (e.g., a digital set top box in a cable network). The system can first operate in a learning mode to receive user inputs and develop evidence that can characterize multiple users of the user equipment device audience. In a working mode, the system can process current user inputs to match a current user to one of the identified users of that user equipment device audience. Fuzzy logic and/or stochastic filtering may be used to improve development of the user characterizations, as well as matching of the current user to those developed characterizations. In this manner, targeting of assets can be implemented not only based on characteristics of a household but based on a current user within that household.

Description

    CROSS-REFERENCE TO RELATED APPLICATIONS
  • This application is a continuation-in-part of U.S. patent application Ser. No. 13/447,071, entitled: “METHOD AND APPARATUS TO PERFORM REAL-TIME ESTIMATION AND COMMERCIAL SELECTION SUITABLE FOR TARGETED ADVERTISING,” filed on Apr. 13, 2012, which is a continuation of U.S. patent application Ser. No. 11/944,078, entitled: “METHOD AND APPARATUS TO PERFORM REAL-TIME ESTIMATION AND COMMERCIAL SELECTION SUITABLE FOR TARGETED ADVERTISING,” filed on Nov. 21, 2007, now abandoned. This application is also a continuation-in-part of U.S. patent application Ser. No. 12/758,532, entitled: “FUZZY LOGIC BASED VIEWER IDENTIFICATION FOR TARGETED ASSET DELIVERY SYSTEM,” filed on Apr. 12, 2010, which is a continuation of U.S. patent application Ser. No. 11/743,544, entitled: “FUZZY LOGIC BASED VIEWER IDENTIFICATION FOR TARGETED ASSET DELIVERY SYSTEM,” filed on May 2, 2007, now U.S. Pat. No. 7,698,236, which claims priority to U.S. Provisional Patent Application No. 60/746,244, entitled: “METHOD AND APPARATUS TO PERFORM REAL-TIME ESTIMATION AND COMMERCIAL SELECTION SUITABLE FOR TARGETED ADVERTISING,” filed on May 2, 2006 and U.S. Provisional Patent Application No. 60/746,245, entitled: “METHOD AND A SYSTEM FOR DISCOVERING THE INDIVIDUAL VIEWERS OF A TELEVISION AUDIENCE (ALONG WITH THEIR ATTRIBUTES, AND BEHAVIOR) AND USING THIS INFORMATION TO ACCURATELY SELECT AND INSERT CANDIDATE ADS IN REAL-TIME FOR ACTIVELY WATCHING VIEWERS,” filed on May 2, 2006. The content of all of these applications is incorporated herein by reference as if set forth in full.
  • FIELD OF INVENTION
  • The present invention relates generally to targeted delivery of assets, such as advertisements or other content, in a communications network. In particular, the invention relates to identifying a current network user and matching assets to the user.
  • BACKGROUND OF THE INVENTION
  • Broadcast network content or programming is commonly provided in conjunction with associated informational content or assets. These assets include advertisements, associated programming, public-service announcements, ad tags, trailers, weather or emergency notifications and a variety of other content, including paid and unpaid content. In this regard, asset providers (e.g., advertisers) who wish to convey information (e.g., advertisements or “ads”) regarding services and/or products to users of the broadcast network often pay for the right to insert their information into programming of the broadcast network. For instance, advertisers may provide ad content to a network operator such that the ad content may be interleaved with broadcast network programming during one or more programming breaks. The delivery of such paid assets often subsidizes or covers the costs of the programming provided by the broadcast network. This may reduce or eliminate costs borne by the users of the broadcast network programming.
  • In order to achieve a better return on their investment, asset providers often try to target their assets to a selected audience that is deemed likely to be interested in the goods or services of the asset provider. The case of advertisers on a cable television network is illustrative. For instance, an advertiser or a cable television network may target its ads to certain demographic groups based on, for example, geographic location, gender, age, income etc. Accordingly, once an advertiser has created an ad that is targeted to a desired group of viewers (e.g., targeted group) the advertiser may attempt to procure insertion times in the network programming when the targeted group is expected to be among the audience of the network programming.
  • More recently, it has been proposed to target assets to individual households. This would allow asset providers to better target audience segments of interest or to tailor messages to different audience segments. However, targeting households is problematic. Again, the case of a cable television network is illustrative. It is often possible to obtain audience classification information for a household based on name or address information. For example, information based on credit card transactions or other financial transactions may be available from third party databases. However, information based on an identified household does not always ensure appropriate targeting of assets. In the case of a family household, for example, a current network user might be a mother, a father, a child, a babysitter, etc. Additionally, where the matching of ads to households is performed in the network, some mechanism is required to target the selected ads to the appropriate households. This is difficult in broadcast networks. Accordingly, household-based targeting, while an improvement over untargeted asset delivery or conventional ratings-based asset targeting in a broadcast network, still entails significant obstacles and/or targeting uncertainty.
  • Current systems have generally suffered from one or more of the following drawbacks: 1) they focus on who is in the household rather than who is watching now; 2) they may only provide coarse information about a subset of the household; 3) they require user participation, which is undesirable for certain users and may entail error; 4) they do not provide a framework for determining when there are multiple viewers or for accurately defining demographics in multiple viewer scenarios; 5) they are fairly static in their assumptions and do not properly handle changing household compositions and demographics; and/or 6) they employ sub-optimal technologies, require extensive training, require excessive resources or otherwise have limited practical application.
  • SUMMARY OF THE INVENTION
  • It has been recognized that the effectiveness of asset targeting can be enhanced by identifying a current network user, e.g., determining demographic or other classification parameters of a putative current network user or users. This would ideally allow an asset targeting system to distinguish between different potential users of a single household, as well as identifying unknown users, such that appropriate targeting of assets can be executed.
  • The present invention enables such functionality in the context of asset delivery in communications networks, including cable television networks. Moreover, such functionality can be executed transparently, from the perspective of the network user, based on monitoring ordinary network usage activities, for example, as indicated by a click stream of a remote control. Moreover, the present invention allows such functionality to be implemented in substantially real-time, using limited processing resources. Thus, for example, the user identification functionality can be executed by an application running on a conventional digital set top box. The invention also provides a mechanism for signaling the network in relation to the user identification process, for example, to enhance selection of assets for insertion into network content streams or to report information for evaluating size and composition of the audience actually reached.
  • Broadly, disclosed herein are methods, systems, and the like (e.g., utilities) for use in targeting assets to users of user equipment devices (e.g., digital set top boxes (DSTBs)) in a communications network. One utility includes operating a processor to progressively incorporate, over time, a plurality of user inputs by one or more users of a user equipment device into a model of a user composition of the one or more users made up of a plurality of user classification parameters; filtering the user composition model to obtain an estimate of a current user composition of the user equipment device (e.g., where the current user composition may be made up of plurality of users); and targeting one or more assets in the communications network using the estimated current user composition.
  • In one arrangement, the filtered user composition model may be free of user equipment device usage patterns for the users obtained before the progressive incorporation of the plurality of user inputs. In this regard, the disclosed utilities can advantageously estimate a current user composition in real time in a manner substantially free of considering previously stored information (e.g., historical periodicity data for one more users). In accordance with another aspect of the present invention, a current user of a communications network can be identified without requiring persistent storage of user profiles. As an example, the utility may include receiving first user inputs at a first time and then receiving second user inputs at a second time after the first time, where the second user inputs are incorporated into the model to a greater degree than are the first user inputs. For instance, a reference event (e.g., disconnection of power to the user equipment device) may be established, where the progressively incorporated inputs occurred at times after the reference event (e.g., and not before the reference event).
  • In this regard, the utility may advantageously incorporate a certain degree of “forgetfulness” therein so as to allow the utility to adapt to change (e.g., from a first user to multiple users to a second user), identify both known and unknown users, limit the amount of viewing information that is available in the system at any one time to address privacy concerns, and the like. Moreover, the utility can be adapted to quickly converge on classification parameters based on contemporaneous user inputs such that errors due to user changes are reduced. An associated utility involves developing a model of a network user based on user inputs free from persistent storage of a profile of the user and using the model of the network user in targeting assets to the user. For instance, recent user inputs may be analyzed using machine learning (e.g., involving fuzzy logic) to determine classification parameters of a current user.
  • In another arrangement, the targeting of assets may include receiving one or more lists of assets for delivery at the user equipment device; obtaining one or more targeting parameters for the one or more lists of assets; and determining a level of correspondence between the user classification parameters of the current user composition of the user equipment device and the targeting parameters for the one or more lists of assets. For instance, the determining may be performed in multiple dimensions relating to multiple classification and targeting parameters. Additionally or alternatively, the utility may include voting for at least one asset of the one or more lists of assets based on the determined level of correspondence between the user classification parameters of the current user composition of the user equipment device and the targeting parameters for the one or more lists of assets. For instance, the at least one voted-for asset may include targeting parameters having a higher level of correspondence with the user classification parameters of the current user composition that does a non-voted for asset.
  • The voting may include sending, from the user equipment device, an indication at least one asset of the lists of assets having targeting parameters with a higher level of correspondence with the user classification parameters of the current user composition that does at least one other asset of the lists of assets. Furthermore, the utility may further include receiving the one or more lists of assets for delivery at a plurality of additional user equipment devices; determining a level of correspondence between user classification parameters of one or more current user compositions of the plurality of additional user equipment devices and the targeting parameters for the one or more lists of assets; voting for at least one asset of the one or more lists of assets based on the determined level of correspondence between the user classification parameters of the current user compositions of the plurality of additional user equipment devices and the targeting parameters for the one or more lists of assets; and aggregating the votes of the user equipment device and the plurality of additional user equipment devices. For instance, the utility may further include receiving one or more updated lists of assets for delivery at the user equipment device and the plurality of additional user equipment devices.
  • In another arrangement, the utility may include progressively incorporating, over time, a plurality of user inputs by users of a plurality of additional user equipment devices into a plurality of models of user composition of the users made up of a plurality of user classification parameters; filtering the user composition models to obtain estimates of current user compositions of the plurality of additional user equipment devices; and aggregating the current user compositions of the user equipment device and the plurality of additional user equipment devices to obtain a current user composition of an aggregated audience, where the targeting includes using the aggregated audience current user composition for use in targeting one or more assets in the communications network. For instance, a level of correspondence between the user classification parameters of the aggregated audience current user composition and the targeting parameters may be determined for the one or more lists of assets.
  • The user composition model may be filtered to obtain current user composition estimates in various manners. For instance, certain aspects of the invention relate to processing corrupted, distorted and/or partial data observations received from the measurement device to infer information about a signal of interest, such as a substantially real time estimate of the state of the signal (e.g., current user composition) at a time of interest. In particular, such a filter system can provide practical approximations of optimized nonlinear filter solutions based on certain constraints on allowable states or combinations therefore inferred from the observation environment.
  • In accordance with one aspect of the present invention, a method and apparatus (“system”) is provided for developing an observation model with respect to data or measurements obtained from the device under analysis. In particular, the system models the input measurements as a Markov chain, whose transitions depend upon the signal. The observation model may take into account exogenous information or information external to (though not necessarily independent of) the input measurements. In one implementation, the input measurements reflect a click stream of DSTB. The click stream may reflect channel selection events and/or other inputs, e.g., related to volume control. In this case, the observation model may further involve programming information (e.g., downloaded from a network platform such as a Head End) associated with selected channels. In this case, it is the click stream information that is processed as a Markov chain.
  • Desired information related to the device can then be obtained by estimating the state of the signal at a time of interest. In the example of analyzing a click stream of a DSTB, the signal may represent a user composition (involving one or more users and/or associated demographics) and an additional factor affecting the click stream such as a channel changing regime as discussed in more detail below. Once the signal has been estimated, a state of the signal at a past, present or future time can be determined, e.g., to provide user composition information for use in connection with an asset targeting system.
  • In accordance with a still further aspect of the present invention, a system generates substantially real time estimates of the probability distribution for a signal state based on both the observations and an observation signal model. In this regard, a nonlinear filter system can be used to provide an estimate of the signal based on the observation model. The nonlinear filter system may involve a nonlinear filter model and an approximation filter for approximating an optimal nonlinear filter solution. For example, the approximation filter may include a particle filter or a discrete state filter for enabling substantially real time estimates of the signal based on the observation model. In the DSTB example, the nonlinear filter system allows for estimates that incorporate user compositions including more than one viewer and adapting to changes in the potential audience, e.g., additions of previously unknown persons or departures of prior users with respect to the potential audience.
  • In accordance with a further aspect of the present invention, a system uses an estimate obtained by applying a filter, with its associated signal and observation models, to a sequence of observations to obtain information of interest with respect to the signal. Specifically, information for a past, present or future time can be obtained based on an estimated probability distribution of the signal at the time of interest. In the case of analyzing usage of a DSTB, the identity and/or demographics of a user or users of the DSTB at a particular time can be determined from the signal state. This information may be used, for example, to “vote” or identify appropriate assets for an upcoming commercial or programming spot, to select an asset from among asset options for delivery at the DSTB and/or to determine or report a goodness of fit of a delivered asset with respect to the user or users who received the asset.
  • In one embodiment of the present invention, a system is provided for use in targeting assets to users of user equipment devices in a communications network, for example, a cable television network. The system involves: developing an observation model based on inputs (e.g., click stream data) by one or more users with respect to a user equipment device (e.g., a DSTB); modeling the signal as reflective of at least a user composition of one or more users of said user equipment device with respect to time; determining the likelihood of various user compositions at a time of interest among possible states of the signal; and using the estimated user composition in targeting an asset for the user equipment device. In this manner, filtering theory is applied with respect to inputs, such as a click stream, of a user equipment device so as to yield an estimate indicative of user composition.
  • The observations (e.g., the inputs) can be modeled as a Markov chain. The model of the signal allows for representation of the user composition as including two or more users. Accordingly, multiple user situations can be identified for use in targeting assets and/or better evaluating audience size and composition (e.g., to improve valuation and billing for asset delivery). In addition, the signal model preferably allows for representation of a change in user composition, e.g., addition or removal of a person from a user audience.
  • A nonlinear filter may be defined to estimate the signal based on the observation model. In this regard, the signal may model the user composition of a household with respect to time and audience classification parameters (e.g., demographics of one or more current users) can be estimated as a function of the state of the signal at a time of interest. In order to provide a practical estimation of an optimal nonlinear filter solution, an approximation filter may be provided for approximating the operation of the nonlinear filter. For example, the approximation filter may include a particle filter or a discrete space filter as described below. Moreover, the approximation filter may implement at least one constraint with respect to one or more signal components. In this regard, the constraint may operate to treat one component of the signal as invariant with respect to a time period where a second component is allowed to vary. Moreover, the constraint may operate to treat at least one state of a first component as illegitimate or to treat some combination of states of different signal components as illegitimate. For example, in the case of a click stream of a DSTB, the occurrence of a click event indicates the certain presence of at least one person. Accordingly, only user compositions corresponding to the presence of at least one person are permissible at the time of a click event. Other permissible or impermissible combinations may relate incomes to locations. The constraints may be implemented in connection with a finite space approximation filter. For example, values incident on an illegitimate cell may be repositioned, e.g., proportionately moved to neighboring legitimate cells. In this manner, the approximation filter can quickly converge on a legitimate solution without requiring undue processing resources. Where the constraint operates to define at least one potential calculated state as illegitimate, the approximation filter may redistribute one or more counts associated therewith.
  • Additionally, the approximation filter may be operative to inhibit convergence on an illegitimate state. Thus, the approximation filter is designed to avoid convergence on a user composition for a DSTB that is logically impossible or unlikely (a click event when no user is present) or deemed illegitimate by rule (an income range not permitted for a given location). In one implementation, this is accomplished by adding seed counts to legitimate cells of a discrete space filter to inhibit convergence with respect to an illegitimate cell.
  • Preferably, the user composition information is processed at the DSTB. That is, user information is processed at the DSTB and used for voting, asset selection and/or reporting. Alternatively, click stream data may be directed to a separate platform, such as a Head End, where the user composition information can be estimated, e.g., where messaging bandwidth is sufficient and DSTB processing resources are limited. As a further alternative, the user composition information (as opposed to, e.g., asset vote information) may be transmitted to a Head End or other platform for use in selecting content for insertion.
  • The estimated user composition information may be used by an asset targeting system. For example, the information may be provided to a network platform such as a Head End that is operative to insert assets into a content stream of the network. In this regard, the platform may utilize inputs from multiple DSTBs to select assets for insertion into available network bandwidth. Additional information, such as information reflecting the per user value of asset delivery, may be utilized in this regard. The platform may process information from multiple user equipment devices as an observation model and apply an appropriately configured filter with respect to the observation model to estimate an overall composition of a network audience at a time of interest.
  • In accordance with another aspect of the present invention, stochastic control theory is applied to the problem of asset selection, e.g., selecting the optimal set of commercial assets to communicate through a limited number of advertising insertion channels. Traditionally, stochastic control theory has been applied in contexts where the state of a system is randomly (time) varying and possibly the exact consequences of various controls applied to the system are only known probabilistically.
  • When one only has noisy, imperfect observations of the system, one must base the set of controls on filtering estimates which are also randomly varying over time. When there are nonlinearities present there is no separation principle to rely on and one must work on a sample path by sample path basis. In the present invention, we do not even get noisy, imperfect observations of the state of the system we want to estimate (i.e., the demographics of the viewers of the various DSTBs), but rather only a noisy partial measurement of the DSTBs estimates of their viewers. Hence, we take the novel approach of designing our system to estimate the set of conditional probability distributions of the DSTBs, from which audience estimates can be obtained as a two-step procedure. We adapt our stochastic control procedures to handle this more general setting.
  • In the present context, sampled viewer estimates from DSTBs received at the Head End are taken to be observations of the system of probability distributions over household viewing states, of arriving advertising contracts, and of ad sale and delivery, in order to allow control decisions regarding which contracts with advertisers to accept. Stochastic control is used to optimize some utility function of the system, e.g., stable profitability.
  • Another manner of filtering the user composition model to obtain the current user composition estimates may involve utilizing machine learning (e.g., fuzzy logic including fuzzy sets and/or fuzzy rules) as part of matching assets to current users. The machine learning system may involve identifying classification parameters of at least one (e.g., current) user based on evidence aggregated from user inputs collected in a learning mode. These inputs may, for example, be analyzed based on correlated programming information, or based on programming independent characteristics, e.g., volume settings or quickness of the click process.
  • For example, fuzzy logic may be used to identify a number of discrete users in an audience (e.g., number of members of a household) and/or to determine one or more classification parameters of a user or users. This may be based on user inputs such as a click stream of a remote control. Thus, user inputs may be monitored and associated with values related to the classification parameter(s). These values can then be treated as points in a fuzzy set within a model of a plurality of user classification parameters. In one implementation, a number of user inputs may be monitored to aggregate points in a fuzzy set. The current user composition estimation and/or matching may involve multiple dimensions related to multiple classification parameters (e.g., age, gender, income, etc.), and the aggregated points may be used to define one or more features of a multidimensional feature terrain. The feature terrain may be processed to remove noise and to reduce the set of gradients in the terrain, for example, by clustering features. The remaining features of the processed feature terrain can then be used to identify each user in an audience and determine one or more classification parameters for each user. Similar processing can be used to identify viewing patterns as a function of time (periodicity). For example, different terrains can be developed for different time periods, e.g., different times of day.
  • Additionally or alternatively, fuzzy logic may be used to develop a characterization of a target audience of a network programming event. For example, the target audience of an asset may be defined by a demographic profile including a number of demographic parameter values. These values may be associated with a series of fuzzy numbers or fuzzy sets. An additional implementation of fuzzy logic may be used to correlate the fuzzy numbers with classification parameters of putative user. For example, a congruent similarity function may be used to match the audience characterization or targeting parameters to the classification parameters. Similar processing can be used to match a periodicity pattern to an identified user. Alternatively, where different terrains are developed for different times, as noted above, such periodicity is reflected in the terrains; that is, time becomes a dimension of the terrain set. A match may be determined based on a combination of the degree of correlation of the user classification parameters to the ad targeting parameters and the likelihood that an appropriate viewer will be watching at the time of the ad delivery. The resulting match may be used to “vote” for assets to be inserted into content streams of the network to select ads for delivery and/or to report a “goodness of fit” of a user receiving the asset to the asset targeting parameters. The noted utility may also be operative to determine whether the user equipment device is “on” and to determine whether any user is present at the user equipment device. The user inputs or click stream data can be processed at the user equipment device or at another location, e.g., raw or preprocessed click stream data may be transmitted to a head end for processing to determine classification parameter information. For example, this may be done where messaging bandwidth is sufficient and user equipment device resources are limited.
  • In one implementation, the machine learning system may be a substantially unsupervised system. That is, the system can accumulate evidence and thereby learn a composition of a user set, such as a viewing audience, without requiring a training process in which the system is provided knowledge about or examples of usage (e.g., viewing) patterns. In this manner, the system can readily adapt to changes, e.g., changes in the viewing audience or viewing audience demographics due to, for example, additions to or departures from the household, changing demographics due to aging, change of income, etc., addition of a television set (e.g., in a child's room) that impacts viewership, etc. Moreover, the system can operate substantially autonomously, thereby substantially avoiding the need for any supervised set-up or retraining process.
  • In accordance with another aspect of the present invention, functionality for identifying a user can be executed at a user equipment device. It has been recognized that a current user can be effectively identified based on analysis of user inputs at a user equipment device. An associated utility in accordance with the present invention involves receiving user inputs at the user equipment device and analyzing the inputs to associate audience classification parameters with the user using a machine learning system. For example, the inputs may relate to a click stream of a remote control device reflecting program selections, volume control inputs and the like. The machine learning system is preferably capable of learning in a substantially unsupervised fashion. Fuzzy logic can be used to analyze these inputs on an individual basis to obtain evidence concerning the classification parameters of the user. This evidence can then be aggregated and analyzed using fuzzy logic to determine classification parameters of a user.
  • In accordance with another aspect of the present invention, a user equipment device is operative to signal a broadcast network regarding a user of the device. An associated utility involves determining, at the user equipment device, user information regarding the user of the device based at least in part on user inputs to the device, and signaling the broadcast network based on user information. For example, the user information may include classification parameters of the user. The signals transmitted to the broadcast network may reflect the results of a matching process whereby user classification information is compared to targeting information for an asset. In this regard, the information transmitted across the network need not include any classification information regarding the user. Such signaling information may be used, for example, to vote for assets to be inserted into network content streams or to report information regarding assets actually delivered at the user equipment device, e.g., for measuring the size and/or composition of the audience.
  • The various aspects of the invention can be provided in any suitable combination. Moreover, any or all of the above noted aspects can be implemented in connection with a targeted asset delivery system.
  • BRIEF DESCRIPTION OF THE DRAWINGS
  • For a more complete understanding of the present invention and further advantages thereof, reference is now made to the following detailed description, taken in conjunction with the drawings in which:
  • FIG. 1 illustrates delivery of assets to different users watching the same programming channel.
  • FIG. 2 illustrates audience aggregation across multiple programming networks.
  • FIG. 3 illustrates a virtual channel in the context of audience aggregation.
  • FIG. 4 illustrates targeted asset insertion being implemented at Customer Premises Equipment (CPEs).
  • FIG. 5 illustrates asset options being transmitted from a headend on separate asset channels.
  • FIG. 6 illustrates a messaging sequence between a CPE, a network platform, and a traffic and billing (T&B) system.
  • FIG. 7A illustrates an example of CPEs that include a television set and a Digital Set Top Box (DSTB) as used by a plurality of users.
  • FIG. 7B illustrates a user classifier.
  • FIG. 8 is a flow chart illustrating a process for implementing time-slot and targeted impression buys.
  • FIG. 9 illustrates an overview of a classifier process in accordance with the present invention.
  • FIG. 10 is a stay transition graph illustrating a process for handling click stream data in accordance with the present invention.
  • FIGS. 11-21 illustrate learning mode operation of the classifier in accordance with the present invention.
  • FIGS. 22-26 illustrate working mode operation of the classifier in accordance with the present invention.
  • FIG. 27 is a block diagram illustrating the basic functional components of a classifier in accordance with the present invention.
  • FIG. 28 is a schematic diagram of a targeted advertising system in accordance with one embodiment.
  • FIG. 29 illustrates the REST structure in accordance with one embodiment.
  • FIG. 30 illustrates a cell structure for a cell of a discrete space filter in accordance with one embodiment.
  • FIG. 31 is a flowchart illustrating a filter evolution process in accordance with one embodiment.
  • FIG. 32 is a block diagram illustrating a process for simulating events in accordance with one embodiment.
  • DETAILED DESCRIPTION
  • The present invention relates to various structure and functionality for delivery of targeted assets, classification of network users, matching of asset targeting parameters to audience classification parameters and network monitoring for use in a communications network. The invention has particular application with respect to networks where content is broadcast to network users. In this regard, content may be broadcast in a variety of networks including, for example, cable and satellite television networks, satellite radio networks, IP networks used for multicasting content and networks used for podcasts or telephony broadcasts/multicasts. Content may also be broadcast over the airwaves though, as will be understood from the description below, certain aspects of the invention make use of bi-directional communication channels which are not readily available, for example, in connection with conventional airwave based televisions or radios (i.e., such communication would involve supplemental communication systems). In various contexts, the content may be consumed in real time or stored for subsequent consumption. Thus, while specific examples are provided below in the context of a cable television network for purposes of illustration, it will be appreciated that the invention is not limited to such contexts but, rather, has application to a variety of networks and transmission modes.
  • The targeted assets may include any type of asset that is desired to be targeted to network users. It is noted that such targeted assets are sometimes referred to as “addressable” assets (though, as will be understood from the description below, targeting can be accomplished without addressing in a point-to-point sense). For example, these targeted assets may include advertisements, internal marketing (e.g., information about network promotions, scheduling or upcoming events), public service announcements, weather or emergency information, or programming. The targeted assets may be independent or included in a content stream with other assets such as untargeted network programming. In the latter case, the targeted assets may be interspersed with untargeted programming (e.g., provided during programming breaks) or may otherwise be combined with the programming as by being superimposed on a screen portion in the case of video programming. In the description below, specific examples are provided in the context of targeted assets provided during breaks in television programming. While this is an important commercial implementation of the invention, it will be appreciated that the invention has broader application. Thus, distinctions below between “programming” and “assets” such as advertising should not be understood as limiting the types of content that may be targeted or the contexts in which such content may be provided.
  • As noted above, the present invention relates to identifying members of an audience, determining classification information for those users, determining which user or users may be watching at a time of interest, and matching assets to the identified audience. The matching related functionality is useful in a variety of contexts in a targeted asset delivery system. Accordingly, an overview of the targeted asset delivery system is first provided below. Thereafter, the matching related functionality and associated structure is described in detail.
  • I. An Exemplary Targeted Asset Delivery System
  • A. The Targeted Asset Delivery Environment
  • Although the matching-related subject matter of the present invention can be used in a variety of targeted asset delivery systems, a particularly advantageous targeted asset delivery system is described below. The inventive system, in the embodiments described below, allows for delivery of targeted assets such as advertising for use in addressing certain shortcomings or inefficiencies of conventional broadcast networks. Generally, such targeting entails delivering assets to desired groups of individuals or individuals having desired characteristics. These characteristics or audience classification parameters may be defined based on personal information, demographic information, psychographic information, geographic information, or any other information that may be relevant to an asset provider in identifying a target audience. Preferably, such targeting is program independent in recognition that programming is a highly imperfect mechanism for targeting of assets. For example, even if user analysis indicates that a particular program has an audience comprised sixty percent of women, and women comprise the target audience for a particular asset, airing on that program will result in a forty percent mismatch. That is, forty percent of the users potentially reached may not be of interest to the asset provider and pricing may be based only on sixty percent of the total audience. Moreover, ideally, targeted asset delivery would allow for targeting with a range of granularities including very fine granularities. For example, it may be desired to target a group, such as based on a geographical grouping, a household characterization or even an individual user characterization. The present invention accommodates program independent targeting, targeting with a high degree of granularity and targeting based on a variety of different audience classifications.
  • FIGS. 1 and 2 illustrate two different contexts of targeted asset delivery supported in accordance with the present invention. Specifically, FIG. 1 illustrates the delivery of different assets, in this case ads, to different users watching the same programming channel, which may be referred to as spot optimization. As shown, three different users 500-502 are depicted as watching the same programming, in this case, denoted “Movie of the Week.” At a given break 504 the users 500-502 each receive a different asset package. Specifically, user 500 receives a digital music player ad and a movie promo, user 501 receives a luxury car ad and a health insurance ad, and user 502 receives a minivan ad and a department store ad. Alternately, a single asset provider (e.g., a motor vehicle company) may purchase a spot and then provide different asset options for the spot (e.g., sports car, minivans, pickup trucks, etc.). Similarly, separate advertisers may collectively purchase a spot and then provide ads for their respective products (e.g., where the target audiences of the advertisers are complementary). It will be appreciated that these different asset packages may be targeted to different audience demographics. In this manner, assets are better tailored to particular viewers of a given program who may fall into different demographic groups. Thus, spot optimization refers to the delivery of different assets (by one or multiple asset providers) in a given spot.
  • FIG. 2 illustrates a different context of the present invention, which may be termed audience aggregation. In this case, three different users 600-602 viewing different programs associated with different channels may receive the same asset or asset package. In this case, each of the users 600-602 receives a package including a digital music player ad and a movie promo in connection with breaks associated with their respective channels. Though the users 600-602 are shown as receiving the same asset package for purposes of illustration, it is likely that different users will receive different combinations of assets due to differences in classification parameters. In this manner, users over multiple channels (some or all users of each channel) can be aggregated (relative to a given asset and time window) to define a virtual channel having significant user numbers matching a targeted audience classification. Among other things, such audience aggregation allows for the possibility of aggregating users over a number of low share channels to define a significant asset delivery opportunity, perhaps on the order of that associated with one of the high share networks. This can be accomplished, in accordance with the present invention, using equipment already at a user's premises (i.e., an existing CPE). Such a virtual channel is graphically illustrated in FIG. 3, though this illustration is not based on actual numbers. Thus, audience aggregation refers to the delivery of the same asset in different spots to define an aggregated audience. These different spots may occur within a time window corresponding to overlapping (conflicting) programs on different channels. In this manner, it is likely that these spots, even if at different times within the window, will not be received by the same users.
  • Such targeting including both spot optimization and audience aggregation can be implemented using a variety of architectures in accordance with the present invention. Thus, for example, as illustrated in FIG. 4, targeted asset insertion can be implemented at the CPEs. This may involve a forward-and-store functionality. As illustrated in FIG. 4, the CPE 800 receives a programming stream 802 and an asset delivery stream 804 from the headend 808. These streams 802 and 804 may be provided via a common signal link such as a coaxial cable or via separate communications links. For example, the asset delivery stream 804 may be transmitted to the CPE 800 via a designated segment, e.g., a dedicated frequency range, of the available bandwidth or via a programming channel that is opportunistically available for asset delivery, e.g., when it is otherwise off air. The asset delivery stream 804 may be provided on a continuous or intermittent basis and may be provided concurrently with the programming stream 802. In the illustrated example, the programming stream 802 is processed by a program-decoding unit, such as DSTB, and programming is displayed on television set 814. Alternatively, the programming stream 802 may be stored in programming storage 815 for CPE insertion.
  • In the illustrated implementation, the asset, together with metadata identifying, for example, any audience classification parameters of the targeted audience, is stored in a designated storage space 806 of the CPE 800. It will be appreciated that substantial storage at the CPE 800 may be required in this regard. For example, such storage may be available in connection with certain digital video recorder (DVR) units. A selector 810 is implemented as a processor running logic on the CPE 800. The selector 810 functions analogously to the headend selector described above to identify breaks 816 and insert appropriate assets. In this case, the assets may be selected based on classification parameters of the household or, more preferably, a user within the household. Such information may be stored at the CPE 800 or may be determined based on an analysis of viewing habits such as a click stream from a remote control as will be described in more detail below. Certain aspects of the present invention can be implemented in such a CPE insertion environment.
  • In FIG. 5, a different architecture is employed. Specifically, in FIG. 5, asset options transmitted from headend 910 synchronously with a given break on a given channel for which targeted asset options are supported. The CPE 900 includes a channel selector 902, which is operative to switch to an asset channel associated with a desired asset at the beginning of a break and to return to the programming channel at the end of the break. The channel selector 902 may hop between channels (between asset channels or between an asset channel and the programming channel) during a break to select the most appropriate assets. In this regard, logic resident on the CPE 900 controls such hopping to avoid switching to a channel where an asset is already in progress. As described below, this logic can be readily implemented, as the schedule of assets on each asset channel is known. Preferably, all of this is implemented invisibly from the perspective of the user of set 904. The different options may be provided, at least in part, in connection with asset channels 906 or other bandwidth segments (separate from programming channels 908) dedicated for use in providing such options. In addition, certain asset options may be inserted into the current programming channel 908. Associated functionality is described in detail below. The architecture of FIG. 5 has the advantage of not requiring substantial storage resources at the CPE 900 such that it can be immediately implemented on a wide scale basis using equipment that is already in the field.
  • As a further alternative, the determination of which asset to show may be made at the headend. For example, an asset may be selected based on voting as described below, and inserted at the headend into the programming channel without options on other asset channels. This would achieve a degree of targeting but without spot optimization opportunities as described above. Still further, options may be provided on other asset channels, but the selection as between those channels may be determined by the headend. For example, information about a household or user (e.g., brand of car owned, magazines subscribed to, etc.) stored on the headend may be used to match an asset to a household or user. That information, which may be termed “marketing labels,” may be used by the headend to control which asset is selected by the CPE. For example, the CPE may be instructed that it is associated with an “ACME preferred” customer. When an asset is disseminated with ACME preferred metadata, the CPE may be caused to select that asset, thereby overriding (or significantly factoring with) any other audience classification considerations. However, it will be appreciated that such operation may entail certain concerns relating to sensitive information or may compromise audience classification based targeting in other respects.
  • A significant opportunity thus exists to better target users whom asset providers may be willing to pay to reach and to better reach hard-to-reach users. However, a number of challenges remain with respect to achieving these objectives including: how to provide asset options within network bandwidth limitations and without requiring substantial storage requirements and new equipment at the user's premises; how to obtain sufficient information for effective targeting while addressing privacy concerns; how to address a variety of business related issues, such as pricing of asset delivery, resulting from availability of asset options and attendant contingent delivery; and how to operate effectively within the context of existing network structure and systems (e.g., across node filters, using existing traffic and billing systems, etc.).
  • From the foregoing it will be appreciated that various aspects of the invention are applicable in the context of a variety of networks, including broadcast networks. In the following discussion, specific implementations of a targeted asset system are discussed in the context of a cable television network. Though the system enhances viewing for both analog and digital users, certain functionality is conveniently implemented using existing DSTBs. It will be appreciated that, while these represent particularly advantageous and commercially valuable implementations, the invention is not limited to these specific implementations or network contexts.
  • B. System Architecture
  • In one implementation, the system of the present invention involves the transmission of asset options in time alignment or synchronization with other assets on a programming channel, where the asset options are at least partially provided via separate bandwidth segments, e.g. channels at least temporarily dedicated to targeted asset delivery. Although such options may typically be transmitted in alignment with a break in programming, it may be desired to provide options opposite continuing programming (e.g., so that only subscribers in a specified geographic area get a weather announcement, an emergency announcement, election results or other local information while others get uninterrupted programming). Selection as between the available options is implemented at the user's premises, as by a DSTB in this implementation. In this manner, asset options are made available for better targeting, without the requirement for substantial storage resources or equipment upgrades at the user's premises (e.g., as might be required for a forward-and-store architecture). Indeed, existing DSTBs can be configured to execute logic for implementing the system described below by downloading and/or preloading appropriate logic.
  • Because asset options are synchronously transmitted in this implementation, it is desirable to be efficient in identifying available bandwidth and in using that bandwidth. Various functionality for improved bandwidth identification, e.g., identifying bandwidth that is opportunistically available in relation to a node filter, is described later in this discussion. Efficient use of available bandwidth involves both optimizing the duty cycle or asset density of an available bandwidth segment (i.e., how much time, of the time a bandwidth segment is available for use in transmitting asset options, is the segment actually used for transmitting options) and the value of the options transmitted. The former factor is addressed, among other things, by improved scheduling of targeted asset delivery on the asset channels in relation to scheduled breaks of the programming channels.
  • The latter factor is addressed in part by populating the available bandwidth spots with assets that are most desired based on current network conditions. These most desired assets can be determined in a variety of ways including based on conventional ratings. In the specific implementation described below, the most desired assets are determined via a process herein termed voting. FIG. 6 illustrates an associated messaging sequence 1000 in this regard as between a CPE 1002 such as a DSTB, a network platform for asset insertion such as a headend 1004 and a traffic and billing (T&B) system 1006 used in the illustrated example for obtaining asset delivery orders or contracts and billing for asset delivery. It will be appreciated that the functionality of the T&B system 1006 may be split between multiple systems running on multiple platforms and the T&B system 1006 may be operated by the network operator or may be separately operated.
  • The illustrated sequence begins by loading contract information 1008 from the T&B system 1006 onto the headend 1004. An interface associated with system 1006 allows asset providers to execute contracts for dissemination of assets based on traditional time-slot buys (for a given program or given time on a given network) or based on a certain audience classification information (e.g., desired demographics, psychographics, geography, and/or audience size). In the latter case, the asset provider or network may identify audience classification information associated with a target audience. The system 1006 uses this information to compile the contract information 1008, which identifies the asset that is to be delivered together with delivery parameters regarding when and to whom the asset is to be delivered.
  • The illustrated headend 1004 uses the contract information together with a schedule of breaks for individual networks to compile an asset option list 1010 on a channel-by-channel and break-by-break basis. That is, the list 1010 lists the universe of asset options that are available for voting purposes for a given break on a given programming channel together with associated metadata identifying the target audience for the asset, e.g., based on audience classification information. The transmitted list 1010 may encompass all supported programming channels and may be transmitted to all participating users, or the list may be limited to one or a subset of the supported channels e.g., based on an input indicating the current channel or the most likely or frequent channels used by a particular user or group of users. The list 1010 is transmitted from the headend 1004 to the CPE 1002 in advance of a break for which options are listed.
  • Based on the list 1010, the CPE 1002 submits a vote 1012 back to the headend 1004. More specifically, the CPE 1002 first identifies the classification parameters for the current user(s) and perhaps the current channel being watched, identifies the assets that are available for an upcoming break (for the current channel or multiple channels) as well as the target audience for those assets and determines a “fit” of one or more of those asset options to the current classification. In one implementation, each of the assets is attributed a fit score for the user(s), e.g., based on a comparison of the audience classification parameters of the asset to the putative audience classification parameters of the current user(s). This may involve how well an individual user classification parameter matches a corresponding target audience parameter and/or how many of the target audience parameters are matched by the user's classification parameters. Based on these fit scores, the CPE 1002 issues the vote 1012 indicating the most appropriate asset(s). Any suitable information can be used to provide this indication. For example, all scores for all available asset options (for the current channel or multiple channels) may be included in the vote 1012. Alternatively, the vote 1012 may identify a subset of one or more options selected or deselected by the CPE 1002, with or without scoring information indicating a degree of the match and may further include channel information. In one implementation, the headend 1004 instructs CPEs (1002) to return fit scores for the top N asset options for a given spot, where N is dynamically configurable based on any relevant factor such as network traffic levels and size of the audience. Preferably, this voting occurs shortly before the break at issue such that the voting more accurately reflects the current status of network users. In one implementation, votes are only submitted for the programming channel to which the CPE is set, and votes are submitted periodically, e.g., every fifteen minutes.
  • The headend 1004 compiles votes 1012 from CPEs 1002 to determine a set of selected asset options 1014 for a given break on a supported programming channel. As will be understood from the description below, such votes 1012 may be obtained from all relevant and participating CPEs 1002 (who may be representative of a larger audience including analog or otherwise non-participating users) or a statistical sampling thereof. In addition, the headend 1004 determines the amount of bandwidth, e.g., the number of dedicated asset option channels, that is available for transmission of options in support of a given break for a given programming channel.
  • Based on all of this information, the headend 1004 assembles a flotilla of assets, e.g., the asset options having the highest vote values or the highest weighted vote values where such weighting takes into account value per user or other information beyond classification fit. Such a flotilla may include asset options inserted on the current programming channel as well as on asset channels, though different insertion processes and components may be involved for programming channel and asset channel insertion. It will be appreciated that some assets may be assembled independently or largely independently of voting, for example, certain public service spots or where a certain provider has paid a premium for guaranteed delivery. Also, in spot optimization contexts where a single asset provider buys a spot and then provides multiple asset options for that spot, voting may be unnecessary (though voting may still be used to select the options).
  • In one implementation, the flotilla is assembled into sets of asset options for each dedicated asset channel, where the time length of each set matches the length of the break, such that channel hopping within a break is unnecessary. Alternatively, the CPE 1002 may navigate between the asset channels to access desired assets within a break (provided that asset starts on the relevant asset channels are synchronized). However, it will be appreciated that the flotilla matrix (where columns include options for a given spot and rows correspond to channels) need not be rectangular. Stated differently, some channels may be used to provide asset options for only a portion of the break, i.e., may be used at the start of the break for one or more spots but are not available for the entire break, or may only be used after one or more spots of a break have aired. A list of the selected assets 1014 and the associated asset channels is then transmitted together with metadata identifying the target audience in the illustrated implementation. It will be appreciated that it may be unnecessary to include the metadata at this step if the CPE 1002 has retained the asset option list 1010. This list 1014 is preferably transmitted shortly in advance of transmission of the asset 1016 (which includes sets of asset options for each dedicated contact options channel used to support, at least in part, the break at issue).
  • The CPE 1002 receives the list of selected asset options 1014 and associated metadata and selects which of the available options to deliver to the user(s). For example, this may involve a comparison of the current audience classification parameter values (which may or may not be the same as those used for purposes of voting) to the metadata associated with each of the asset options. The selected asset option is used to selectively switch the CPE 1002 to the corresponding dedicated asset options channel to display the selected asset 1016 at the beginning of the break at issue. One of the asset option sets, for example, the one comprised of the asset receiving the highest vote values, may be inserted into the programming channel so that switching is not required for many users. Assuming that the voting CPEs are at least somewhat representative of the universe of all users, a significant degree of targeting is thereby achieved even for analog or otherwise non-participating users. In this regard, the voters serve as proxies for non-voting users. The CPE 1002 returns to the programming channel at the conclusion of the break. Preferably, all of this is transparent from the perspective of the user(s), i.e., preferably no user input is required. The system may be designed so that any user input overrides the targeting system. For example, if the user changes channels during a break, the change will be implemented as if the targeting system was not in effect (e.g., a command to advance to the next channel will set the CPE to the channel immediately above the current programming channel, without regard to any options currently available for that channel, regardless of the dedicated asset channel that is currently sourcing the television output).
  • In this system architecture, as in forward-and-store architectures or any other option where selections between asset options are implemented at the CPE, there will be some uncertainty as to how many users or households received any particular asset option in the absence of reporting. This may be tolerable from a business perspective. In the absence of reporting, the audience size may be estimated based on voting data, conventional ratings analysis and other tools. Indeed, in the conventional asset delivery paradigm, asset providers accept Nielsen rating estimates and demographic information together with market analysis to gauge return on investment. However, this uncertainty is less than optimal in any asset delivery environment and may be particularly problematic in the context of audience aggregation across multiple programming networks, potentially including programming networks that are difficult to measure by conventional means.
  • The system of the present invention preferably implements a reporting system by which individual CPEs 1002 report back to the headend 1004 what asset or assets were delivered at the CPE 1002 and, optionally, to whom (in terms of audience classification). Additionally, the reports may indicate where (on what programming channel) the asset was delivered and how much (if any) of the asset was consumed. Such reports 1018 may be provided by all participating CPEs 1002 or by a statistical sampling thereof. These reports 1018 may be generated on a break-by-break basis, periodically (e.g., every 15 minutes) or may be aggregated prior to transmission to the headend 1004. Reports may be transmitted soon after delivery of the assets at issue or may be accumulated, e.g., for transmission at a time of day where messaging bandwidth is more available. Moreover, such reporting may be coordinated as between the CPEs 1002 so as to spread the messaging load due to reporting.
  • In any case, the reports 1018 can be used to provide billing information 1020 to the T&B system 1006 for valuing the delivery of the various asset options. For example, the billing information 1020 can be used by the T&B system 1006 to determine how large an audience received each option and how well that audience matched the target audience. For example, as noted above, a fit score may be generated for particular asset options based on a comparison of the audience classification to the target audience. This score may be on any scale, e.g., 1-100. Goodness of fit may be determined based on this raw score or based on characterization of this score such as “excellent,” “good,” etc. Again, this may depend on how well an individual audience classification parameter of a user matches a corresponding target audience parameter and/or how many of the target audience parameters are matched by the user's audience classification parameters. This information may in turn be provided to the asset provider, at least in an aggregated form. In this manner, the network operator can bill based on guaranteed delivery of targeted messages or scale the billing rate (or increase delivery) based on goodness of fit as well as audience size. The reports (and/or votes) 1018 can also provide a quick and detailed measurement of user distribution over the network that can be used to accurately gauge ratings, share, demographics of audiences and the like. Moreover, this information can be used to provide future audience estimation information 1022, for example, to estimate the total target universe based on audience classification parameters.
  • It will thus be appreciated that the present invention allows a network operator such as an MSO to sell asset delivery under the conventional asset delivery (time-slot) buy paradigm or under the new commercial impression paradigm or both. For example, a particular MSO may choose to sell asset delivery space for the major networks (or for these networks during prime time) under the old time-slot buy paradigm while using the commercial impression paradigm to aggregate users over multiple low market share networks. Another MSO may choose to retain the basic time-slot buy paradigm while accommodating asset providers who may wish to fill a given slot with multiple options targeted to different demographics. Another MSO may choose to retain the basic time-slot buy paradigm during prime time across all networks while using the targeted impression paradigm to aggregate users at other times of the day. The targeted impression paradigm may be used by such MSOs only for this limited purpose.
  • FIG. 8 is a flow chart illustrating an associated process 1200. An asset provider (or agent thereof) can initiate the illustrated process 1200 by accessing (1202) a contracting platform as will be described below. Alternatively, an asset provider can work with the sales department or other personnel of a system operator or other party who accesses such a platform. As a still further alternative, an automated buying system may be employed to interface with such a platform via a system-to-system interface. This platform may provide a graphical user interface by which an asset provider can design a dissemination strategy and enter into a corresponding contract for dissemination of an asset. The asset provider can then use the interface to select (1204) to execute either a time-slot buy strategy or a targeted impression buy strategy. In the case of a time-slot buy strategy, the asset provider can then use the user interface to specify (1206) a network and time-slot or other program parameter identifying the desired air times and frequency for delivery of the asset. Thus, for example, an asset provider may elect to air the asset in connection with specifically identified programs believed to have an appropriate audience. In addition, the asset provider may specify that the asset is to appear during the first break or during multiple breaks during the program. The asset provider may further specify that the asset is to be, for example, aired during the first spot within the break, the last spot within the break or otherwise designate the specific asset delivery slot.
  • Once the time-slots for the asset have thus been specified, the MSO causes the asset to be embedded (1208) into the specified programming channel asset stream. The asset is then available to be consumed by all users of the programming channel. The MSO then bills (1210) the asset provider, typically based on associated ratings information. For example, the billing rate may be established in advance based on previous rating information for the program in question, or the best available ratings information for the particular airing of the program may be used to bill the asset provider. It will thus be appreciated that the conventional time-slot buy paradigm is limited to delivery to all users for a particular time-slot on a particular network and does not allow for targeting of particular users of a given network or targeting users distributed over multiple networks in a single buy.
  • In the case of targeted impression buys, the asset provider can use the user interface as described in more detail below to specify (1212) audience classification and other dissemination parameters. In the case of audience classification parameters, the asset provider may specify the gender, age range, income range, geographical location, lifestyle interest or other information of a targeted audience. The additional dissemination parameters may relate to delivery time, frequency, audience size, or any other information useful to define a target audience. Combinations of parameters may also be specified. For example, an asset provider may specify an audience size of 100,000 in a particular demographic group and further specify that the asset is not delivered to any user who has already received the asset a predetermined number of times.
  • Based on this information, the targeted asset system of the present invention is operative to target appropriate users. For example, this may involve targeting only selected users of a major network. Additionally or alternatively, this may involve aggregating (1214) users across multiple networks to satisfy the audience specifications. For example, selected users from multiple programming channels may receive the asset within a designated time period in order to provide an audience of the desired size, where the audience is composed of users matching the desired audience classification. The user interface preferably estimates the target universe based on the audience classification and dissemination parameters such that the asset provider receives an indication of the likely audience size.
  • The aggregation system may also be used to do time of day buys. For example, an asset provider could specify audience classification parameters for a target audience and further specify a time and channel for airing of the asset. CPEs tuned to that channel can then select the asset based on the voting process as described herein. Also, asset providers may designate audience classification parameters and a run time or time range, but not the programming channel. In this manner, significant flexibility is enabled for designing a dissemination strategy. It is also possible for a network operator to disable some of these strategy options, e.g., for business reasons.
  • Based on this input information, the targeted asset system of the present invention is operative to provide the asset as an option during one or more time-slots of one or more breaks. In the case of spot optimization, multiple asset options may be disseminated together with information identifying the target audience so that the most appropriate asset can be delivered at individual CPEs. In the case of audience aggregation, the asset may be provided as an option in connection with multiple breaks on multiple programming channels. The system then receives and processes (1218) reports regarding actual delivery of the asset by CPEs and information indicating how well the actual audience fit the classification parameters of the target audience. The asset provider can then be billed (1220) based on guaranteed delivery and goodness of fit based on actual report information. It will thus be appreciated that a new asset delivery paradigm is defined by which assets are targeted to specific users rather than being associated with particular programs. This enables both better targeting of individual users for a given program and improved reach to target users on low-share networks.
  • From the foregoing, it will be appreciated that various steps in the messaging sequence are directed to matching assets to users based on classification parameters, allowing for goodness of fit determinations based on such matching or otherwise depending on communicating audience classification information across the network. It is preferable to implement such messaging in a manner that is respectful of user privacy concerns and relevant regulatory regimes.
  • In the illustrated system, this is addressed by implementing the system free from persistent storage of a user profile or other sensitive information including, for example, personally identifiable information (PII). Specifically, it may be desired to protect as sensitive information subject matter extending beyond the established definition of PII. As one example in this regard, it may be desired to protect MAC addresses even though such addresses are not presently considered to be included within the definition of PII in the United States. Generally, any information that may entail privacy concerns or identify network usage information may be considered sensitive information. More particularly, the system learns of current network conditions prior to transmission of asset options via votes that identify assets without any sensitive information. Reports may also be limited to identifying assets that have been delivered (which assets are associated with target audience parameters) or characterization of the fit of audience classification parameters of a user(s) to a target audience definition. Even if it is desired to associate reports with particular users, e.g., to account for ad skipping as discussed below, such association may be based on an identification code or address not including PII. In any event, identification codes or any other information deemed sensitive can be immediately stripped and discarded or hashed, and audience classification information can be used only in anonymous and aggregated form to address any privacy concerns. With regard to hashing, sensitive information such as a MAC or IP address (which may be included in a designated header field) can be run through a hash function and reattached to the header, for example, to enable anonymous identification of messages from the same origin as may be desired. Moreover, users can be notified of the targeted asset system and allowed to opt in or opt out such that participating users have positively assented to participate.
  • Much of the discussion above has referenced audience classification parameters as relating to individuals as opposed to households. FIG. 7A illustrates a theoretical example of a CPE including a television set 1100 and a DSTB 1102 that are associated with multiple users 1103-1106. Arrow 1107 represents a user input stream, such as a click stream from a remote control, over time. A first user 1105, in this case a child, uses the television 1100 during a first time period—for example, in the morning. Second and third users 1103 and 1104 (designated “father” and “mother”) use the television during time periods 1109 and 1110, which may be, for example, in the afternoon or evening. A babysitter 1106 uses the television during a nighttime period in this example.
  • This illustrates a number of challenges related to targeted asset delivery. First, because there are multiple users 1103-1106, targeting based on household demographics would have limited effectiveness. For example, it may be assumed that the child 1105 and father 1103 in many cases would not be targeted by the same asset providers. Moreover, in some cases, multiple users may watch the same television at the same time as indicated by the overlap of time periods 1109-1110. In addition, in some cases such as illustrated by the babysitter 1106 an unexpected user (from the perspective of the targeted asset system) may use the television 1100.
  • These noted difficulties are associated with a number of objectives that are preferably addressed by the targeted asset system of the present invention. First, the system should preferably be operative to distinguish between multiple users of a single set and, in the context of the system described above, vote and report to the network accordingly. Second, the system should preferably react over time to changing conditions such as the transitions from use by father 1103 to use by both father and mother 1103 and 1104 to use by only mother 1104. The system should also preferably have some ability to characterize unexpected users such as the babysitter 1106. In that case, the system may have no other information to go on other than the click stream 1107. The system may also identify time periods where, apparently, no user is present, though the set 1100 may still be on. Preferably, the system also operates free from persistent storage of any user profile or sensitive information so that no third party has a meaningful opportunity to misappropriate such information or discover the private network usage patterns of any of the users 1103-1106 via the targeted asset system. Privacy concerns can alternatively be addressed by obtaining consent from users. In this matter, sensitive information including PII can be transmitted across the network and persistently stored for use in targeting. This may allow for compiling a detailed user profile, e.g., at the headend. Assets can then be selected based on the user profile and, in certain implementations, addressed to specific CPEs.
  • In certain implementations, the present invention monitors the click stream over a time window and applies a mathematical model to match a pattern defined by the click stream to predefined audience classification parameters that may relate to demographic or psychographic categories. It will be appreciated that the click stream will indicate programs selected by users, volume and other information that may have some correlation, at least in a statistical sense, to the classification parameters. In addition, factors such as the frequency of channel changes and the length of time that the user lingers on a particular asset may be relevant to determining a value of an audience classification parameter. The system can also identify instances where there is apparently no user present.
  • In a first implementation, as is described in detail below, logic associated with the CPE 1101 uses probabilistic modeling, fuzzy logic and/or machine learning to progressively estimate the audience classification parameter values of a current user or users based on the click stream 1107. This process may optionally be supplemental based on stored information (preferably free of sensitive information) concerning the household that may, for example, affect probabilities associated with particular inputs. In this manner, each user input event (which involves one or more items of change of status and/or duration information) can be used to update a current estimate of the audience classification parameters based on associated probability values. The fuzzy logic may involve fuzzy data sets and probabilistic algorithms that accommodate estimations based on inputs of varying and limited predictive value.
  • In a second implementation, the click stream is modeled as an incomplete or noisy signal that can be processed to obtain audience classification parameter information. More specifically, a series of clicks over time or associated information can be viewed as a time-based signal. This input signal is assumed to reflect a desired signature or pattern that can be correlated to audience classification parameters. However, the signal is assumed to be incomplete or noisy—a common problem in signal processing. Accordingly, filtering techniques are employed to estimate the “true” signal from the input stream and associated algorithms correlate that signal to the desired audience classification information. For example, a nonlinear adaptive filter may be used in this regard. Modeling user inputs (e.g., a clickstream) as an incomplete/noisy signal and then utilizing various filtering techniques to estimate a state of the signal to obtain a current audience classification (e.g., in terms of classification parameters) will be discussed in more detail later in this discussion.
  • In any event, certain preferred characteristics apply in either of these noted examples. First, the inputs into the system are primarily a click stream and stored aggregated or statistical data, substantially free of any sensitive information. This addresses privacy concerns as noted above but also provides substantial flexibility to assess new environments such as unexpected users. In addition, the system preferably has a forgetfulness such that recent inputs are more important than older inputs. Either of the noted examples accommodates this objective. It will be appreciated that such forgetfulness allows the system to adapt to change, e.g., from a first user to multiple users to a second user. In addition, such forgetfulness limits the amount of viewing information that is available in the system at any one time, thereby further addressing privacy concerns, and limits the time period during which such information could conceivably be discovered. For example, information may be deleted and settings may be reset to default values periodically, for example, when the DSTB is unplugged.
  • A block diagram of a system implementing such a user classification system is shown in FIG. 7B. The illustrated system is implemented in a CPE 1120 including a user input module 1122 and a classification module 1124. The user input module receives user inputs, e.g., from a remote control or television control buttons, that may indicate channel selections, volume settings and the like. These inputs are used together with programming information 1132 (which allows for correlation of channel selections to programming and/or associated audience profiles) for a number of functions. In this regard, the presence detector 1126 determines whether it is likely that a user is present for all or a portion of an asset that is delivered. For example, a long time period without any user inputs may indicate that no user is present and paying attention or a volume setting of zero may indicate that the asset was not effectively delivered. The classifier 1128 develops audience classification parameters for one or more users of a household as discussed above. The user identifier is operative to estimate which user, of the classified users, is currently present. Together, these modules 1126, 1128 and 1130 provide audience classification information that can be used to vote (or elect not to vote) and/or generate reports (or elect not to generate reports).
  • As noted above, one of the audience classifications that may be used for targeting is location. Specifically, an asset provider may wish to target only users within a defined geographic zone (e.g., proximate to a business outlet) or may wish to target different assets to different geographic zones (e.g., targeting different car ads to users having different supposed income levels based on location). In certain implementations, the present invention determines the location of a particular CPE and uses the location information to target assets to the particular CPE. It will be appreciated that an indication of the location of a CPE contains information that may be considered sensitive. The present invention also creates, extracts and/or receives the location information in a manner that addresses these privacy concerns. This may also be accomplished by generalizing or otherwise filtering out sensitive information from the location information sent across the network. This may be accomplished by providing filtering or sorting features at the CPE or at the headend. For example, information that may be useful in the reporting process (i.e. to determine the number of successful deliveries within a specified location zone) may be sent upstream with little or no sensitive information included. Additionally, such location information can be generalized so as to not be personally identifiable. For example, all users on a given block or within another geographic zone (such as associated with a zip plus 2 area) may be associated with the same location identifier (e.g., a centroid for the zone).
  • In one implementation, logic associated with the CPE sends an identifier upstream to the headend where the identifier is cross-referenced against a list of billing addresses. The billing address that matches the identifier is then translated, for example, using GIS information, into a set of coordinates (e.g., Cartesian geographic coordinates) and those coordinates or an associated geographic zone identifier are sent back to the CPE for storage as part of its location information. Alternatively, a list may be broadcast. In this case, a list including location information for multiple or all network users is broadcast and each CPE selects its own information. Asset providers can also associate target location information with an asset. For example, in connection with a contract interface as specified below, asset providers can define target asset delivery zones. Preferably this can be done via a graphical interface (e.g., displaying a map), and the defined zones can match, to a fine level of granularity, targeted areas of interest without being limited to node areas or other network topology. Moreover, such zones can have complex shapes including discontiguous portions. Preferably the zones can then be expressed in terms that allow for convenient transmission in asset metadata and comparison to user locations e.g., in terms of grid elements or area cells.
  • In another implementation, individual geographic regions are associated with unique identifiers and new regions can be defined based on the union of existing regions. This can be extended to a granularity identifying individual CPEs at its most fine level. Higher levels including numerous CPEs may be used for voting and reporting to address privacy concerns.
  • Upon receipt of an asset option list or an asset delivery request (ADR), the CPE parses the ADR and determines whether the location of the CPE is included in the locations targeted by the asset referenced in the ADR. For example, this may involve a point in polygon or other point in area algorithm, a radius analysis, or a comparison to a network of defined grid or cells such as a quadtree data structure. The CPE may then vote for assets to be received based on criteria including whether the location of that particular CPE is targeted by the asset.
  • After displaying an asset option, the CPE may also use its location information in the reporting process to enhance the delivery data sent upstream. The process by which the CPE uses its location information removes substantially all sensitive information from the location information. For example, the CPE may report that an asset targeted to a particular group of locations was delivered to one of the locations in the group. The CPE in this example would not report the location to which asset was actually delivered.
  • Similarly, it is often desired to associate tags with asset selections. Such tags are additional information that is superimposed on or appended to such assets. For example, a tag may provide information regarding a local store or other business location at the conclusion of an asset that is distributed on a broader basis. Conventionally, such tags have been appended to ads prior to insertion at the headend and have been limited to coarse targeting. In accordance with the present invention, tags may be targeted to users in particular zones, locations or areas, such as neighborhoods. Tags may also be targeted based on other audience classification parameters such as age, gender, income level, etc. For example, tags at the end of a department store ad may advertise specials on particular items of interest to particular demographics. Specifically, a tag may be included in an asset flotilla and conditionally inserted based on logic contained within the CPE. Thus the tags are separate units that can be targeted like other assets, however, with conditional logic such that they are associated with the corresponding asset.
  • The present invention may use information relating to the location of a particular CPE to target a tag to a particular CPE. For example, the CPE may contain information relating to its location in the form of Cartesian coordinates as discussed above. If an asset indicates that a tag may be delivered with it or instead of it, the CPE determines whether there is, associated with any of the potential tags, a location criterion that is met by the location information contained in the particular CPE. For example, a tag may include a location criterion defining a particular neighborhood. If the CPE is located in that neighborhood, the CPE 1101 may choose to deliver the tag, assuming that other criteria necessary for the delivery of the tag are met. Other criteria may include the time available in the given break, other demographic information, and information relating to the national or non-localized asset.
  • As briefly note above, targeting may also be implemented based on marketing labels. Specifically, the headend may acquire information or marketing labels regarding a user or household from a variety of sources. These marketing labels may indicate that a user buys expensive cars, is a male, is 18-24 years old, and/or include other information of potential interest to an asset provider. In some cases, this information may be similar to the audience classification parameters, though it may optionally be static (not varying as television users change) and based on hard data (as opposed to being surmised based on viewing patterns or the like). In other cases, the marketing labels may be more specific or otherwise different than the audience classification. In any event, the headend may inform the CPE as to what kind of user/household it is in terms of marketing labels. An asset provider can then target an asset based on the marketing labels and the asset will be delivered by CPEs where targeting matches. This can be used in audience aggregation and spot optimization contexts.
  • Thus, the targeted asset system of the present invention allows for targeting of assets in a broadcast network based on any relevant audience classification, whether determined based on user inputs such as a click stream, based on marketing labels or other information pushed to the customer premises equipment, based on demographic or other information stored or processed at the headend, or based on combinations of the above or other information. In this regard, it is therefore possible to use, in the context of a broadcast network, targeting concepts that have previously been limited to other contexts such as direct mail. For example, such targeting may make use of financial information, previous purchase information, periodical subscription information and the like. Moreover, classification systems developed in other contexts, may be leveraged to enhance the value of targeting achieved in accordance with the present invention.
  • An overview of the system has thus been provided, including introductory discussions of major components of the system, which provides a system context for understanding the operation of the matching related functionality and associated structure. This matching related subject matter is described in the remainder of this description.
  • II. Asset Matching
  • A. Overview
  • From the discussion above, it will be appreciated that determining classification parameters for a user and matching the classification parameters of the user to targeting parameters of an asset is useful in several contexts. First, this matching-related functionality is useful in the voting process. That is, one of the functions of the targeting system in the system described above is to receive ad lists (identifying a set of ads that are available for an upcoming spot), determining the targeting parameters for the various ads and voting for one or more ads based on how well the targeting parameters match the classification parameters of a current users. Thus, identifying the classification parameters of the current user(s) and matching those parameters to the targeting parameters is important in the voting context.
  • Matching related functionality is also important in the ad selection context. Specifically, after the votes from the various participating set top boxes have been processed, a flotilla of ads is assembled for a commercial break. A given DSTB selects a path through the flotilla (corresponding to a set of ads delivered by the set top box at the commercial break) based on which ads are appropriate for the user(s). Accordingly, identifying the current user classification parameters and matching those parameters to the targeting parameters is important in the ad selection context.
  • The matching related functionality may also be used in the reporting context. In this regard, some or all at the DSTBs provide reports to the network concerning the ads that were actually delivered. This enables the targeting system and the traffic and billing system to measure the audience for an ad so that the advertiser can be billed appropriately. Preferably, the information provided by these reports not only indicates the size of the audience but how well the audience fits the target audience for the ad. Accordingly, the system described above can provide goodness of fit information identifying how well the classification parameters of the user(s) who received the ad match the targeting parameters. The matching related functionality is also useful in this context.
  • In one implementation of the present invention, this matching related functionality is performed by a classifier resident at the DSTB. This classifier will be described in more detail below. Generally, the classifier analyzes a click stream, or series of remote control inputs, to determine probable classification parameters of a current user(s). The classifier also performs a matching function to determine a suitability of each of multiple candidate ads (e.g., from an ad list) for the current user(s) based on inferred classification parameters of a putative user or users. The classifier can then provide matching related information for use in the voting, ad selection and reporting contexts as described above. In the description below, this matching related information is primarily discussed in relation to the voting context. However, it will be appreciated that corresponding information can be used for ad selection and reporting. In addition, while the invention can be used in connection with targeting various types of assets in various networks, the following description is set forth in the context of targeting ads in a cable television network. Accordingly, the terms “ad” and “viewer” are used for convenience and clarity. Moreover, for convenience, though the classifier can identify multiple users, the description below sometimes refers to a singular user or viewer.
  • The classifier generally operates in two modes: the learning mode and the working mode. It will be appreciated, however, that these modes are not fully separate. For example, the classifier continues to learn and adapt during normal operation. These modes are generally illustrated in FIG. 9. In the learning mode, the illustrated classifier 1300 monitors behavior of viewers 1302 in the audience of a given DSTB 1304 to deduce classification parameters for the viewers 1302. In this regard, an audience for a given DSTB 1304 may include a father, a mother and a child, one or more of whom may be present during a viewing session. The classification parameters may include any of the classification parameters noted above, such as gender, age, income, program preferences or the like.
  • As shown in FIG. 9, in the learning mode, the classifier receives inputs including click data 1310 from the user, program data 1312 (such as program guide data) from the network and Nielsen data 1314 generated by the Nielsen system. This information is processed to learn certain behaviors of the viewer, including a viewer program selection behavior. In this regard, the Nielsen data 1314 reflects the demographic composition for particular programs. The program data 1312 may include information regarding the genre, rating, scheduled time, channel and other information regarding programs. The click data 1310 reflects channels selected by a user, dwell time (how long a user remained on a given channel) and other information that can be correlated with the Nielsen data 1314 and program data 1312 to obtain evidence regarding classification parameters of the viewer. In addition, the click data 1310 may reflect frequency of channel hopping, quickness of the click process, volume control selections and other information from which evidence of viewer classification parameters may be inferred independent of any correlation to program related information. As will be discussed in more detail below, in the learning mode, the classifier 1300 may begin to generate clusters of data around segments of classification parameter values (e.g., associated with conventional data groupings), thereby learning to identify viewers 1310 and classify the viewers 1302 in relation to their probable gender, age, income and other classification parameters.
  • The state transition functionality is illustrated in FIG. 10. State changes are triggered by events, messages and transactions. One of the important state transitions is the stream of click events 1400. Each click event 1400 represents a state transition (for example, a change from one program to another, a change in volume setting, etc.). As shown, an absence of click events 1400 or low frequency of click events may indicate that no viewer is present or that any viewing is only passive 1404. If the transition count 1406 exceeds a threshold (e.g., in terms of frequency) and the DSTB is on or active 1402, then the classifier may be operated to match program data 1408 and learn or store 1410 the program. Programs may be deleted 1412 in this regard so as not to exceed a maximum stack depth or to implement a degree of desirable forgetfulness as described above.
  • Referring again to FIG. 9, as the learning mode progresses, viewer identifications are developed in relation to at least two sets of characteristic information. First, a classification parameter set is developed for each discovered viewer 1302 of the DSTB audience. Second, for each discovered viewer 1302, a set of rules is developed that defines the viewing behavior over time for that viewer 1302. This is referred to below as the periodicity of the viewer's viewing habits. Thus, the classification set for each discovered viewer 1302 may identify the viewer's age, gender, education, income and other classification parameters. This information is coupled with the periodicity of the viewer's viewing habits so as to allow the classifier to match an ad with a target audience during a specific timeframe. That is, the determination by the classifier as to who is watching at a given time may be informed both by a substantially real-time analysis of viewing behavior and by historical viewing patterns of a viewer 1302. Alternatively, this process of developing classification parameter sets for discovered users may take into consideration multiple time frames, e.g., different times of day. Developing these classification parameter sets for discovered viewers as a function of time of day, e.g., on an hourly, half-hourly or other time dependent basis, has been found effective, as viewership in many households is significantly dependent on time of day.
  • It will be appreciated that the learning mode and working mode need not be distinct. For example, the classifier may estimate classification parameters for a current viewer 1302 even if historical periodicity data has not been developed for that viewer 1302. Similarly, even where such information has been developed in the learning mode, a current viewing audience may be continually defined and redefined during the working mode. Thus, the classifier does not require persistent storage of viewer profile information in order to function. For example, any such stored information may be deleted when the DSTB is turned off. In such a case, the classifier can readily develop classification parameters for one or more viewers when power is restored to the set top box. Moreover, the classifier may be designed to incorporate a degree of forgetfulness. That is, the classifier may optionally de-weight or delete aged information from its evolving model of audience members. In this manner, the classifier can adapt to changes in the audience composition and identify previously unknown audience members.
  • When sufficient viewer behavior information has been collected (which may only require a small number of user inputs), the illustrated classifier moves from the learning mode to the working mode. In the working mode, the classifier 1306 performs a number of related functions. First, it can receive ad lists 1308 for an upcoming commercial break, match the targeting parameters for the ads on that list 1308 to the classification parameters of the current viewer 1302, and vote for appropriate ads. The classifier 1306 also selects ads available in a given ad flotilla for delivery to the current user 1302. Moreover, the classifier 1306 can report goodness of fit information regarding ads delivered during one or more commercial breaks. Again, in the working mode, the classifier 1306 continues to learn through a process of stochastic reinforcement, but the classifier 1306 is deemed to have sufficient information to meaningfully estimate classification parameters of a current viewer 1302 or audience.
  • As noted above, in the working mode, the classifier 1306 controls the voting process by effectively ranking ads from an ad list 1308. In this regard, the illustrated working mode classifier 1306 receives information regarding available ads 1316 from an ad repository 1318. These ads 1316 are associated with targeting parameters, for example, in the form of audience segmentation and viewer profile classification rules 1320. For example, an advertiser may enter targeting parameters directly into the T & B system via the ad interface. Typically, these targeting parameters may be defined in relation to conventional audience segmentation categories. However, as discussed above, the targeting system of the present invention may accommodate different or finer targeting parameters. The working mode classifier 1306 also receives an ad list 1308 or a view list of candidate ads, as described above. Specifically, the headend targeting system component 1320 processes the inputs regarding available ads and their targeting parameters to generate the ad list 1308 for distribution to participating DSTBs.
  • The similarity and proximity analyzer 1322 uses the targeting parameters associated with individual ads and the classification parameters of the current user to execute matching functionality. That is, the analyzer 1322 matches an ad with at least one of the probable viewers 1302 currently thought to be sitting in front of the television set 1324. As will be described in more detail below, this is done by comparing, for example, the target age range for an ad (which may be expressed as a slightly fuzzified region) to the set representing the viewer's age (which may also be a fuzzy set). The more these two sets overlap the greater the compatibility or match. Such matching is performed in multiple dimensions relating to multiple targeting/classification parameters. This similarity analysis is applied across each candidate ad of the ad list 1308, and a degree of similarity is determined for each ad. When this process is complete, the ads in each time period can be sorted, e.g., in descending order by similarity, and one or more of the top ads may be selected for voting.
  • The passive voting agent 1326 is operative to select ads based on the match information. This process works in the background generally using an out of band data stream. More specifically, the illustrated voting agent 1326 selects a record or ADR for each of the candidate ads and determines for each if any viewers are likely to be present at the ad time. Additionally, the voting agent 1326 determines if any such viewer has classification parameters that acceptably match the targeting parameters for the ad. In the voting context, for each match, a vote is made for the ad. This vote is returned to the headend component 1320 where it is combined with other votes. These aggregated votes are used to generate the next generation of ad lists 1308.
  • An overview of the classifier system has thus been provided. The learning mode, working mode and matching functionality is described in more detail in the following sections.
  • B. Learning Mode Operation
  • As noted above, the learning mode classifier develops classification parameter information for probable viewers as well as periodicity information for those viewers. This process is illustrated in more detail in FIG. 11. During learning mode, the classifier 1500 is constructing a statistical model of the audience. In particular, it is desirable to develop a model that enables feature separability—the ability to reliably distinguish between identified viewers of an audience. It is thus desired to have a good definition of the classification parameters of each viewer and the ability to identify the current viewer 1502, from among the identified viewers of the audience, each time the classifier needs to know who is watching the television. In FIG. 11, this process is illustrated with respect to two examples of the classification parameters; namely, target gender and target age.
  • As discussed above, the learning mode classifier 1500 receives inputs including programming viewing frequency (or demographic) data from Nielsen, BBM or another ratings system (or based on previously reported information of the targeted advertising system) 1504, program data 1506 and click data 1508. Based on this data, the classifier 1500 can determine which program, if any, is being viewed at a particular time and what that indicates regarding probable classification parameters of a viewer. Additionally, the click data 1508, since it is an event stream, may indicate the level of focus or concentration of the viewer 1502 at any time and may provide a measure of the level of interest in a particular program. The click data 1508 also allows the classifier 1500 to determine when the DSTB set is turned off and when it is turned on.
  • The learning mode classifier 1500 fuses the incoming data through a set of clustering and partitioning techniques designed to uncover the underlying patterns in the data. The goal in this regard is to discover the number of probable viewers, build a classification parameter set for each viewer and determine the viewing habits of each viewer over time (or to develop classification parameter sets for likely viewers as a function of time). This information allows the classifier 1500 to determine what kind of audience probably exists at the delivery time for a specific ad. In the illustrated example, this learning process involves the development of two classifier modules—the age and gender classifier module 1510 and the viewing behavior or periodicity classifier module 1512. The periodicity classifier 1512 accumulates and reinforces the results of the age and gender classification module across time. A sequence of age interval classes 1510 are stored across an independent axis representing the time of day and day of week that that evidence is collected. This time axis is used to determine the time of day that each individual detected in the age and gender classifier module 1510 tends to watch television.
  • The gender and age classifier module 1510 gathers evidence over time as to probable viewers. Once sufficient evidence is collected, it is expected that the evidence will cluster in ways that indicate a number of separate audience members. This is at once a fairly simple and complex process. It is simple because the core algorithms used to match viewing habits to putative age and gender features are well understood and fairly easy to implement. For example, it is not difficult to associate a program selection with a probability that the viewer falls into certain demographic categories. On the other hand, it is somewhat complex to analyze the interplay among parameters and to handle subtle phenomena associated with the strength or weakness of the incoming signals. In the latter regard, two parameters that affect the learning process are dwell time and Nielsen population size. The dwell time relates to the length of time that a viewer remains on a given program and is used to develop an indication of a level of interest. Thus, dwell time functions like a filter on the click stream events that are used in the training mechanism. For example, one or more thresholds may be set with respect to dwell time to attenuate or exclude data. In this regard, it may be determined that the classifier does not benefit from learning that a viewer watched a program if the viewer watched that program for less than a minute or, perhaps, less than 10 seconds. Thresholds and associated attenuation or exclusion factors may be developed theoretically or empirically in this regard so as to enhance identification accuracy.
  • Also related to dwell time is a factor termed the audience expectation measure. This is the degree to which, at any time, it is expected that the television (when the DSTB is turned on) will have an active audience. That is, it is not necessarily desirable to have the classifier learn what program was tuned in if nobody is in fact watching. The audience expectation measure can be determined in a variety of ways. One simple measure of this factor is the number of continuous shows that has elapsed since the last channel change or other click event. That is, as the length of time between click events increases, the confidence that someone is actively watching decreases. This audience expectation measure can be used to exclude or attenuate data as a factor in developing a viewer identification model.
  • Nielsen marketing research data is also useful as a scaling and rate-of-learning parameter. As noted above, this Nielsen data provides gender and age statistics in relation to particular programs. As will be discussed in more detail below, click events with sufficient dwell time may be used to accumulate evidence with respect to each classification parameter segment, e.g., a fuzzy age interval. In this regard, each piece of evidence effectively increments the developing model such that classification parameter values are integrated over time. How much a given fuzzy parameter set is incremented is a function of the degree of membership that a piece of evidence possesses with respect to each such fuzzy set.
  • Thus, the degree of membership in a particular age group is treated as evidence for that age. However, when the program and time is matched with the Nielsen data, the gender distribution may contain a broad spectrum of viewer population frequencies. The dwellage percentage of the audience that falls into each age group category is also evidence for that age group. Accordingly, the amount that a set is incremented is scaled by the degree of membership with respect to that set. Thus, for example, if few viewers of an age category are watching a program, this is reflected in only a small amount of evidence that the viewer is in this age group.
  • The illustrated learning mode classifier 1500 also encompasses a periodicity classifier module 1512. As the classifier 1500 develops evidence that allows for determining the number of viewers in an audience of a DSTB and for distinguishing between the viewers, it is possible to develop a viewing model with respect to time for each of these viewers. This information can then be used to directly predict who is likely to be watching at the time of ad delivery. There are a number of ways to build the periodicity model, and this can be executed during the learning mode operation and/or the working mode operation. For example, this model may involve mapping a viewer to their pattern. Alternatively, a pattern may be discovered and then matched to a known viewer. In the illustrated implementation, the latter viewer-to-pattern approach is utilized. As will be understood from the description below, this approach works well because the properties that define the periodicity are fuzzy numbers. The match can therefore use the same kind of similarity function that is used to match viewers to targeting parameters of ads.
  • The discussion above noted that viewers are identified based on integrating or aggregating evidence in relation to certain (e.g., fuzzy) sets. This process may be more fully understood by reference to FIG. 12. In this case, which for purposes of illustration is limited to discovery of age and gender, this involves a 2×M fuzzy pattern discovery matrix. The two rows are gender segmentation vectors. The M columns are conventional age intervals used in ad targeting. These age intervals are overlaid with fuzzy interval measures. In the illustrated example, the classifier is modeled around certain age intervals (12-17, 25-35, 35-49) because these are industry standard segmentations. It will be appreciated, however, that specific age groups are not a required feature of the classifier.
  • The fuzzy intervals are represented by the trapezoidal fuzzy set brackets that illustrate a certain amount of overlap between neighboring age intervals. This overlap may improve discrimination as between different age ranges. The bars shown on the matrix reflect the accumulation of evidence based on a series of click events. As can be seen in the matrix of FIG. 12, over time, this evidence tends to cluster in a fashion that indicates discrete, identifiable viewers associated with different classification parameters.
  • This is further illustrated in FIG. 13. In the example of FIG. 13, this process of accumulating evidence to identify discrete viewers of an audience is depicted in three-dimensional graphics. Thus, the result of the learning process is a collection of gradients or hills or mountains in the learning matrix or multi-dimensional (in this case, two-dimensional) feature terrain. In this case, the higher hills or “mountains” with their higher gradient elevations provide the best evidence that their site in the learning matrix is the site of a viewer. The hills are constructed from the counts in each of the cells defined by the classification parameter segmentation (e.g., age groups). The greater the count, the higher the hill and, consequently, the more certain it is that the characteristics correspond to a viewer of the audience. Viewed from above, the gradients and their elevations form a topological map, as shown in FIG. 14. The concentration and height of the contour lines reflect a clustering that suggests discrete and identifiable viewers. Thus, the map of FIG. 14 reflects three probable viewers identified from a learning process. It is noted that the fuzzy terrain mapping, which allows certain degree of overlap between surrounding age groups, provides a more refined estimate of the actual classification parameter values of a viewer. That is, an interpolation of evidence in adjacent fuzzy sets enables an estimation of an actual parameter value that is not limited to the set definitions. This interpolation need not be linear and may, for example, be executed as a center of gravity interpolation. That takes into account gradient height in each fuzzy set as well as a scaled degree of membership of the height in the set.
  • The process for applying evidence to the terrain may thus result in a fuzzified terrain. That is, rather than simply applying a “count” to a cell of the terrain based on determined classification parameter values, a count can effectively be added as terrain feature (e.g., a hill) centered on a cell (or on or near a cell boundary) but including residual values that spill over to adjacent cells. The residual values may spill over onto adjacent cells in multiple dimensions. In one implementation, this effect is defined by a proximity calculation. The result of applying evidence to the terrain in a fuzzified fashion is potentially enhanced user definition as well as enhanced ability to distinguish as between multiple users.
  • The proximity algorithm and terrain seeding noise reduction filtering can be illustrated by reference to an example involving three classification parameter dimensions; namely, gender, age and time of day. The feature terrain may be viewed as being defined by a cube composed of a number of sub-cubes or grains. For example, the terrain cube may be composed of 7680 grains—2×80×48 grains, corresponding to two gender, 80 age and 48 half-hour time categories. Each of the grains can be populated with a reference or noise value. For example, the noise values can be derived from a statistical analysis of the expected viewers at each age, gender and time measured over all available programming channels for which Nielson, BBM or another ratings system have observers (or based on previously reported information of the targeted advertising system). The noise may be drawn from a distribution of these viewers from a suitable function (e.g., developed empirically or theoretically) and may or may not be as simple as an average or weighted average. In any case, it is this noise bias that cancels out a corresponding randomness in the evidence leaving a trace or residue of evidence only when the evidence is repeatedly associated with an actual viewer whose behavior matches the (age, gender, time) coordinates of the terrain.
  • It will be appreciated that the grain definition and population of the grains with reference values or noise is not limited to the granularity of the source of the reference values, e.g., the standard Neilson categories. Rather, the reference values can be interpolated or estimated to match the defined grain size of the terrain cube. Thus, for example, Neilson source data may be provided in relation to 16 age gender groups whereas the terrain cube, as noted above, may include 2×80 corresponding grains (or columns of grains where the column axis corresponds to the time dimension). Various mathematical techniques can be used in this regard. For example, the age distribution can be fitted to a curve or function, which can then be solved for each age value. The corresponding values are then applied to seed the terrain.
  • Evidence is the statistical frequency of viewers of a particular age and gender who are watching a television program (which is playing at a particular time). The combination of Neilson frequencies and the program time generates 16 pieces of evidence for each time period (there are 8 male and 8 female age groups with their viewing frequencies). Finding out which of these 16 pieces of evidence corresponds to the actual viewer is the job of the statistical learning model underlying the classifier. Fundamental to making this decision is the methodology used to add evidence to the terrain. This involves first applying the noted noise filter as follows:

  • d=f(a,g,t)−r(a,g,t)  (1)
  • where (f) is the observed evidence and (r) represents the noise for that age, gender and time. This value (d) will be either positive or negative. The result is added to the terrain (t) on a grain-by-grain basis:

  • t(a,g,t)=t(a,g,t)+d  (2)
  • If the evidence is being drawn randomly from the incoming statistics, then the number of positive and negative residuals (d) will be approximately equal, and the total sum of the terrain value (t) at that point will also be zero. If, however, the evidence is associated with an actual viewer, then we would expect a small but persistent bias in the frequency statistics to accumulate around that real viewer. Over time, this means that residual (d) will be positive more often than negative and that the contour at the terrain cube will begin to grow. As more evidence is added, the contour grows as a small hill on the terrain. This small hill is an actual viewer (actually, because of the time axis, viewers appear as a ridge of connected hills, somewhat like a winding mountain range).
  • But because this is a statistical learning model, and because reinforcement is sporadic (due to inconsistent viewer behaviors), it is useful to fuzzify or spread the evidence, e.g., to surround each emerging contour with a small bit of probabilistic evidence that will help us define an actual viewer's behavior in the time dimension. To do this, we take the evidence (d) and use it to populate adjoining grains. This is done in a series of concentric circles out from the target terrain grain. Thus, for the first set of adjoining grains (in all directions in age, gender and time), we add x1=d*.10, for the next, x2=x1*.05, for the next, x3=x2*.025, etc., until the multiplier fall below some threshold. It will be appreciated that these multipliers are simply examples and other values, derived empirically or theoretically, can be used. As an example, if d=100, then the proximity values would look something like this,
  • 0.0125 0.0125 0.0125 0.0125 0.0125 0.0125 0.0125
    0.0125 0.5 0.5 0.5 0.5 0.5 0.0125
    0.0125 0.5 10 10 10 0.5 0.0125
    0.0125 0.5 10 100 10 0.5 0.0125
    0.0125 0.5 10 10 10 0.5 0.0125
    0.0125 0.5 0.5 0.5 0.5 0.5 0.0125
    0.0125 0.0125 0.0125 0.0125 0.0125 0.0125 0.0125
  • As can be seen, the value at each outward layer is based on a fraction of the previous layer (not the original value). Of course, in a real terrain, the original value is very small, so the values in each outward cell become very, very small very, very quickly. Yet, over time, they provide enough additional evidence to support the growth of a valid contour site.
  • The proximity algorithm that fills out the terrain, also provides a quick and effective way of discovering who is viewing the television associated with the DSTB. The ridge of contours or hills associated with a viewer wanders across the three-dimensional terrain cube. These hills are smoothed out (made “fatter,” so to speak) over the terrain by the addition of the minute bits of partial evidence laid down during the building of the terrain (the x1, x2, x3, etc., in the previous section). To find the viewer, the hills over the gender/age axes at a particular time can be summed and averaged. The viewer is the hill with the maximum average height. Optionally, a proximity algorithm can be used in integrating these hills. For example, when determining the height value associated with a given grain, a fraction of the height of adjacent grains and a small fraction of the next outward layer of grains (etc.) may be added. This is somewhat analogous to the mountain clustering algorithm discussed herein and may be used as an alternative thereto.
  • As noted above, these terrains may be developed as a function of time of day. For example, the data may be deposited in “bins” that collect data from different times of day on an hourly, half-hourly or other basis (e.g., irregular intervals matching morning, daytime, evening news, primetime, late night, etc.). Again, the process of applying the evidence to these bins may be fuzzified such that evidence spills over to some extent into adjacent bins or cells in the time and other dimensions. Evidence may be integrated in these bins over multiple days. The resulting terrains may be conceptualized as multiple terrains corresponding to the separate timeframes or as a single terrain with a time dimension. This functionality may be implemented as an alternative to the separate periodicity analysis described below.
  • The previous examples have reflected relatively clean datasets that were easy to interpret so as to identify discrete viewers and their associated classification parameters. In reality, real world data may be more difficult to interpret. In this regard, FIG. 15 illustrates a more difficult dataset in this regard. In particular, the feature terrain of FIG. 15 does not readily yield interpretation as to the number of viewers in the audience or their specific classification parameters. Any number of factors may result in such complexity. For example, the remote control may be passed between different audience members, the click stream may be influenced by other audience members, a click event of significant duration may reflect distraction rather than interest, a viewer may have a range of programming interests that do not cleanly reflect their classification parameters, etc.
  • In order to resolve complicated data such as illustrated in FIG. 15, the classifier implements processes of noise removal and renormalization of the gradient terrain. The goal is to find the actual centers of evidence so that the number of viewers and their classification parameters can be accurately identified. One type of process that may be implemented for removing or attenuating noise involves consideration of reference values, e.g., average values for all events at that time, taken in relation to the whole audience. For example, the terrain may be seeded with reference values or the reference values may be considered in qualifying or rejecting data corresponding to individual events. In one implementation, data is compared to reference values on an event-by-event basis to qualify data for deposit into the bins for use in developing the terrain. This has the effect of rejecting data deemed likely to represent noise. In effect, the reference values are subtracted from evidence as it is applied to the terrain, thus impeding the process of constructing terrain features so that terrain features substantially only rise as a result of persistent or coherent accumulation of evidence likely reflective of an actual user, and spurious peaks are avoided. Selecting the reference values as average values, weighted average values or some other values related to the observation context (but substantially without information likely to bias the user identification and definition problem) has the effect of scaling the filter effect to properly address noise without inhibiting meaningful terrain construction. Additionally, data may be qualified in relation to a presence detector. As noted above, presence may be indicated by reference to the on/off state of the DSTB and/or by the absence or infrequency of inputs over a period of time. Data acquired when the presence detector indicates that no user is deemed “present” may be excluded.
  • Noise removal may further involve eliminating false centers and sporadic evidence counts without disturbing the actual centers to the extent possible. In this regard, FIG. 16 illustrates a possible identification of viewer sites with respect to the data of FIG. 15. FIG. 17 illustrates an alternative, also potentially valid, interpretation of the data of FIG. 15. The classifier implements an algorithm designed to determine which of competing potentially valid interpretations is most likely correct. This algorithm generally involves gradient deconstruction. The deconstruction process is an iterative process that finds gradient centers by first removing low-level interference noise (thus revealing the candidate hills) and then measuring the compactness of the distribution of hills around the site area. In this regard, a mountain clustering algorithm can be implemented to iteratively identify peaks, remove peaks and revise the terrain. The affect is to identify cluster centers and to smooth out the hilliness between cluster centers so that the cluster centers become increasingly distinct. The remaining sites after the feature terrain is processed in this regard are the sites of the putative viewers.
  • This mountain clustering and noise reduction functionality may be further understood by reference to FIGS. 18-21. FIG. 18 shows a learning matrix or feature terrain with a set of gradients scattered over the surface. In the first iteration of the algorithm, as shown in FIG. 19, gradient C1 is identified as the maximum mountain. The distance from C1's center of gravity to the centers of each gradient on the surface (e.g., C2, C3, etc.) is measured. The degree of mountain deconstruction on each gradient is inversely proportional to the square distance between C1 and each of the other clusters. This inverse square mechanism localizes nearly all of the mountain deconstruction to the neighborhood of C1. Thus, FIG. 20 shows significant deconstruction of features proximate to C1 as a result of this first iteration. This causes the general perimeter of the emerging site at C1 to contract. The contraction also involves the assimilation of small gradients, hence the height of C1 is increased, and the adjacent hilliness is reduced. At the same time, the height of C3 is barely changed and its set of satellite gradients has not yet begun to be assimilated. The process is then repeated with respect to C3. After a number of iterations, the final sites begin to emerge and stabilize. FIG. 21 shows the terrain after most of the smaller gradients have been assimilated. The resulting processed terrain or feature space has well defined gradients at each putative viewer site. The mountain clustering process thus essentially removes much ambiguity.
  • A similar process is performed with respect to the periodicity analysis. Specifically, a periodicity learning matrix is created and updated in a manner analogous to construction and updating of the viewer classification parameter matrix and its conversion to the feature terrain. In this case, the periodicity terrain produces a set of gradients defining, by their height and width, the expectation that the viewer is watching television at that point in time. A matching algorithm can then be used to match a periodicity pattern to one of identified viewers.
  • C. Working Mode Operation
  • When the classifier has been sufficiently trained, it moves from learning mode operation to working mode operation. As noted above, these modes are not entirely distinct. For example, the classifier can perform estimations of classification parameters while still in the learning mode, and the classifier continues to learn during the working mode. However, as described above, the learning mode is a gradual process of collecting evidence and measuring the degree to which the viewer sites are discernable in the learning matrix. Thus while the classifier can operate quickly using default values, working mode operation reflects a determination that the viewers of an audience have been identified and classified with a high degree of confidence.
  • The basic operation of the working mode classifier is illustrated in FIG. 22. In the working mode, the classifier 2600 has access to the developed feature terrain, as discussed above, as well as to a similarity function that is operative to match targeting parameters to the classification parameters as indicated by the viewer identification. More specifically, Nielsen data 2602 and program data 2604 continues to be fed into the classifier 2600 in the working mode to support its continued learning process that essentially runs in the background. The classifier 2600 then receives an ADR from the ad list 2608. The ADR is initially filtered for high level suitability and then, if the ADS is still available for matching analysis, selects each viewer and each classification parameter for each viewer for comparison. Thus, FIG. 22 depicts the process of accessing a classification matrix 2606 of a viewer. In this case, the matrix 2606 is a two-dimensional matrix limited to gender and age classification parameters. In practice, any number of classification parameters may be supported.
  • The retrieved classification parameter values of the user and the targeting parameters of the ad are then provided to the similarity and proximity analyzer 2612 where they are compared. This function returns a degree of similarity for each attribute. The total compatibility or relative compatibility rank (RCR) is given as:
  • RCR = i = l N S i × w i i = l N w i
      • Where,
        • si is the similarity measurement for the ith property
        • wi is the weighting factor for the ith property. If weighting (priority or ranking) is not used, the default, w=1, means that weights do not affect the ranking.
  • Through this process, an ad may be found to be compatible with one or more viewers of an audience. For example, such compatibility may be determined in relation to an RCR threshold value. Thus, when an ad is found to be compatible with one or more of the viewers, the periodicity analyzer 2616 is called to see if the viewer is likely to be present at the target ad insertion time. If the viewer is unlikely to be watching, the degree of this time constraint is used to adjust the RCR. Accordingly, the RCR is recomputed as:
  • R C R = i = l N s i × w i i = l N w i × l ( pT )
      • Where,
        • 1( ) is the likelihood estimating function from the periodicity analysis. This function returns a fuzzy degree of estimate in the interval [0,1] (which is actually a degree of similarity between the target time period and each of the viewer's active time periods)
        • pT is the target time period
  • Accordingly, the RCR will have a high value if there is both a high degree of match between the classification parameters of a user and the targeting parameters of the ad, and there is a high probability that the user will be watching at the ad delivery time. This formula further has the desirable quality that a low compatibility where the viewer is not watching, even if the viewer's classification parameters are a very good match for a given ad.
  • The voting agent 2614 is closely connected to the operation of the working mode classifier 2600. In particular, the validity of the vote is highly dependent on the ability to correctly identify a viewer's classification parameters and viewing habits. Thus, the voting agent 2614 essentially works in the same way as the view list ranking Specifically, an ADR is sent to the voting agent 2614, the voting agent 2614 extracts the ads targeting parameters and calls the classifier 2600, which returns the RCR for this ad. In this manner, ads are not only voted on but also delivered based on a matching process.
  • D. Matching Functionality
  • The similarity function used to execute the various matching functionality as discussed above can be understood by reference to FIGS. 23-26. Thus, as noted above, when an ADR is received at the classifier to be ranked, the targeting parameters and any ad constraints are embedded in the ADR. The ADR may also indicate an “importance” of the ad. For example, such importance may be based on the ad pricing (e.g., CPM value) or another factor (e.g., a network operator may specify a high importance for internal marketing, at least for a specified geographic area such as where a competing network is, or is becoming, available). The classifier also has access to the classification parameters of the various viewers in the audience, as discussed above. The first step in the matching process is applying a similarity function, e.g., a fuzzy similarity function, to each of the classification parameters and targeting parameters. The similarity function then determines the degree to which a targeting parameter is similar to or compatible with a classification parameter. The weighted average of the aggregated similarity of values for each of the classification/targeting parameters is the base match score.
  • More specifically, the illustrated classifier fuzzifies the targeting parameter and finds the degree of membership of the corresponding classification parameter of the user in this fuzzified region. Thus, as shown in FIG. 23, the targeting parameters for an ad may specify a target age range of 24-42 years of age. As shown in FIG. 23, this targeting parameter is redefined as a fuzzy set. Unlike rigid sets, the fuzzy set has a small but real membership value across the entire domain of the classification parameter (age in this case). The membership function means that the matching process cannot automatically identify and rank classification parameters values that lie near but, perhaps, not inside the target range. As an example, FIG. 24 shows a putative viewer with an inferred age of 26. In this case, the viewer sits well inside the target age range. The degree of membership is therefore 1.0, indicating a complete compatibility with the target age. In fact, any age that is within the target age range will return a matching membership of 1.0. However, the matching process can also deal with situations where retrieved classification parameters do not match the targeting parameter range. FIG. 25 illustrates the case where a viewer is estimated to be 20 years old and is therefore outside of the rigid target range boundaries. If the classifier had Boolean selection rules, this viewer would not be selected. The nature of the fuzzy target space, however, means that viewer 2 is assigned a similarity or compatibility value greater than zero, in this case 0.53. The classifier now has the option of including viewer 2, knowing that the viewer is moderately compatible with the target range.
  • FIG. 26 shows a slightly different situation—that of a viewer with an inferred age outside of the required age interval. As illustrated by the membership function, as the viewer age moves away from the identified targeting age interval, the membership function drops off quickly. In this case, viewer 3 has a compatibility of only 0.18, indicating that it would normally be a poor candidate for an ad with a 24-42 age requirement. It is noted, however, that the fuzzy compatibility mechanism means that the classifier can find and rank a viewer if any viewer exists.
  • E. Stochastic Filtering
  • As discussed previously, audience classification parameters (e.g., of a household, of a current audience, etc.) can also be determined or estimated by way of modeling a click stream as an incomplete or noisy signal and processing the signal to obtain such audience classification information. The information obtained via filtering of the noisy signal can be used in a variety of contexts such as for voting, ad selection and/or goodness of fit determinations as described above. Alternatively, the description below describes a filter theory based Head End ad selection system that is an alternative to the noted voting processes. As a still further alternative, click stream information can be provided to the Head End, or another network platform, where the audience classification parameters may be calculated. Thus, the audience classification parameter, ad selection and other functionality can be varied and may be distributed in various ways between the DSTBs, Head End or other platforms.
  • The following section is broken into several parts. In the first part, some background discussion of the relevant nonlinear filter theory is provided. In the second part, the architecture and model classes are discussed.
  • 1.1 Nonlinear Filtering
  • To properly solve the targeted advertisement viewership (potential and current) problem, one may look to the mathematically optimal field of filtering.
  • 1.1.1 Traditional Nonlinear Filtering Overview
  • Nonlinear filtering deals with the optimal estimation of the past, present and/or future state of some nonlinear random dynamic process (typically called ‘the signal’) in real-time based on corrupted, distorted or partial data observations of the signal. In general, the signal Xt is regarded as a Markov process defined on some probability space (Ω, ℑ,P) and is the solution to some Martingale problem. The observations typically occur at discrete times tk and are dependent upon the signal in some stochastic manner using a sensor function Yk=h(Xt k , Vk). Indeed, the traditional theory and methods are built around this type of observations, where the measurements are distorted (by nonlinear function h), corrupted (by noise V), partial (by the possible dependence of h on only part of the signal's state) samples of the signal. The optimal filter provides the conditional distribution of the state of the signal given the observations available up until the current time:

  • P(X t εdx|σ{Y k,0≦t k ≦t})
  • The filter can provide optimal estimates for not only the current states of the signal but for previous and future states, as well as path segments of the signal:

  • P(X [t r ,t s ] εdx|σ{Y k,0≦t k ≦t})
  • where 0≦tr≦ts<∞.
  • In certain linear circumstances, an effective optimal recursive formula is available. Suppose the signal follows a “linear” stochastic differential equation dXt=AXtdt+BdWt, with A being a linear operator, B being a fixed element and W being a Brownian motion. Furthermore, the observation function takes the form of Yk=CXt k +Vk where {Vk}k=1 are independent Gaussian random variables and C is a linear operator. This formula is known as the Kalman filter. While the Kalman filter is very efficient in performing its estimates, its use in applications is inherently limited due to the strict description of the signal and observation processes. In the case where the dynamics of the signal are nonlinear, or the observations have non-additive and/or correlated noise, the Kalman filter provides sub-optimal estimates. As a result, other methods are sought out to provide optimal estimates in these more common scenarios.
  • While equations for optimal nonlinear estimation have been available for several decades, until recently they were found to be of little use. The optimal equations were unimplementable on a computer, requiring infinite memory and computational resources to be used. However, in the past decade and a half, approximations to the optimal filtering equations have been created to overcome this problem. These approximations are typically asymptotically optimal, meaning that as an increasing amount of resources are used in their computation they converge to the optimal solution. The two most prevalent types of such methods are particle methods and discrete space methods.
  • 1.1.2 Particle Filters
  • Particle filtering methods involve creating many copies of the signal (called ‘particles’) denoted as {ξt j}j=1 N 1 , where Nt is the number of particles being used at time t. These particles are evolved independently over time according to the signal's stochastic law. Each particle is then assigned a weight value W1,mt j) to effectively incorporate the information from the sequence of observations {Y1, . . . , Ym}. This can be done in such a way that the weight after m observations is the weight after m−1 multiplied by a factor dependent on the mth observation Ym. However, these weights invariably become extremely uneven meaning that many particles (those with relatively low weights) become unimportant and do little other than consume computer cycles. Rather than only removing these particles and reducing calculation to an ever-decreasing number of particles, one resamples the particles, which means the positions and weights of particles are adjusted to ensure that all particles contribute to the conditional distribution calculation in a meaningful way while ensuring that no statistical bias is introduced by this adjustment. Early particle methods tended to resample far too extensively, introducing excessive resampling noise into the system of particles and degrading estimates. Suppose that after resampling the weights of the particles after m observations are denoted as {tilde over (W)}1,mt j}j=1 N 1 . Then, the particle filter's approximation to the optimal filter's conditional distribution is:
  • P ( X t A Y 1 , , Y m ) j = 1 N t m W ~ 1 , m ( ξ j ) 1 ξ tm j A j = t N t m W ~ 1 , m ( ξ j )
  • As Nt→∞, the particle-filtering estimate yields the optimal nonlinear filter estimate.
  • An improvement that introduced significantly less resampling degradation and improved computational efficiency was introduced in U.S. Pat. No. 7,058,550, entitled “Selectively Resampling Particle Filter,” which is incorporated herein by reference. This method performed pair-wise resampling as follows:
  • 1. While {tilde over (W)}1,mj)<p{tilde over (W)}1,mi) for the highest weighted particle j and the lowest weighted particle i, then:
  • 2. Set the state of particle i to j with probability
  • W ~ 1 , m ( ξ j ) W ~ 1 , m ( ξ j ) + W ~ 1 , m ( ξ i )
  • and set the state of particle j to i with probability
  • W ~ 1 , m ( ξ j ) W ~ 1 , m ( ξ j ) + W ~ 1 , m ( ξ i ) .
  • 3. Reset the weight of particles i and j to
  • W ~ 1 , m ( ξ j ) = W ~ 1 , m ( ξ i ) = W ~ 1 , m ( ξ j ) W ~ 1 , m ( ξ i ) 2 .
  • In this method, a control parameter ρ is introduced to appropriately moderate the amount of resampling performed. As described in U.S. Pat. No. 7,058,550, this value can be dynamic over time in order to adapt to the current state of the filter as well as the particular application. This filing also included efficient systems to store and compute the quantities required in this algorithm on a computer.
  • 1.1.3 Discrete Space Filters
  • When the state space of the signal is on some bounded finite dimensional space, then a discrete space and amplitude approximation can be used. A discrete space filter is described in detail in U.S. Pat. No. 7,188,048, entitled “Refining Stochastic Grid Filter” (REST Filter), which is incorporated herein by reference. In this form, the state space D is partitioned into discrete cells ηc for c in some finite index set C. For instance, this space D could be a d-dimensional Euclidean space or some counting measure space. Each cell yields a discretized amplitude known as a “particle count” (denoted as nηc), which is used to form the conditional distribution of the discrete space filter:
  • P ( X t A Y 1 , , Y m ) c C n η c 1 η c A c C n η c
  • The particle counts of each state cell are altered according to the signal's operator as well as the observation data that is processed. As the number of cells becomes infinite, then the REST filter's estimate converges to the optimal filter. To be clear, this filing considers directly discretizing filtering equations rather than discretizing the signal and working out an implementable filtering equation for the discretized signal.
  • In U.S. Pat. No. 7,188,048, the invention utilized a dynamic interleaved binary index tree to organize the cells with data structures in order to efficiently recursively compute the filter's conditional estimate based on the real-time processing of observations. While this structure was amenable to certain applications, in scenarios where the dimensional complexity of the state space is small, the data structure's overhead can reduce the method's utility.
  • 1.2 Stochastic Control
  • To properly solve the targeted commercial selection problem, one should look to the mathematically optimal field of stochastic control.
  • Conceptually, one could invent particle methods or direct discretization methods to solve a stochastic control problem approximately on a computer. However, these have not yet been implemented or at least widely recognized. Instead, implementation methods usually discretize the whole problem and then solve the discretized problem.
  • 2.1 Targeted Advertising System Architecture FIG. 28 depicts the overall targeted advertising system. The system is composed of a Head End 4100 and one or more DSTBs 4200. The DSTBs 4200 are attempting to estimate the conditional probability of the state of potential viewers in household 4205, including the current member(s) of the household watching television, using the DSTB filter 4202. The DSTB filter 4202 uses a pair of models 4201 describing the signal (household) and the observations (the click stream data 4206). The DSTB filter 4202 is initialized via the setting 4302 downloaded from the Head End 4100. To estimate the state of the household the DSTB filter 4202 also uses program information 4207 (which may be current, or in the recent past or future), which is available from a store of program information 4208.
  • The DSTB filter 4202 passes its conditional distribution or estimates derived thereof to a commercial selection algorithm 4203, which then determines which commercials 4204 to display to the current viewers based on the filter's output, the downloaded commercials 4301, and any rules 4302 that govern what commercials are permissible given the viewer estimates. The commercials displayed to the viewers are recorded and stored.
  • The DSTB filter 4202 estimates, as well as commercial delivery statistics and other information, may be randomly sampled 4303 and aggregated 4304 to provide information to the Head End 4100. This information is used by a Head End filter 4102, which computes (subject to its available resources) the conditional distribution for the aggregate potential and actual viewership for the set of DSTBs with which it is associated. The Head End filter 4102 uses an aggregate household and DSTB feedback model 4101 to provide its estimates. These estimates are used by the Head End commercial selection system 4103 to determine which commercials should be passed to the set of DSTBs controlled by the Head End 4100. The commercial selection system 4103 also takes into account any market information 4105 available concerning the current commercial contracts and economics of those contracts. The resulting commercials selected 4301 are subsequently downloaded to the DSTBs 4200. The commercials selected for downloading affect the level settings 4104, which provide constraints on certain commercials being shown to certain types of individuals.
  • The following two sections describe certain detail elements of this system.
  • 2.2 Household Signal and Observation Model Description In this section, the general signal and observation model description are given as well as examples of possible embodiment of this model.
  • 2.2.1 Signal Model Description
  • In general, the signal of a household is modeled as a collection of individuals and a household regime. In one preferred embodiment, this household represents the people who could potentially watch a particular television that uses a DSTB. Each individual (denoted as Xi) at a given point in time t has a state from the state space sεS, where S represents the set of characteristics that one wishes to determine for each person within a household. For example, in one embodiment one may wish to classify the age, gender, income, and watching status of each individual. In addition, it has been found that certain behavioral information, in particular, the amount of television watched by each individual, is useful in developing and using classifications. Age and income may be considered as real values, or as a discrete range. In this example, the state space would be defined as:

  • S={0-12, 12-18, 18-24, 24-38, 38+}×{Male, Female}×{0-$50,000, $50,000+}×{Yes, No}
  • The household member tuple is then
  • k = 0 S k ,
  • where k denotes the number of individuals and S0 denotes the single state with no individuals. The household member tuple Xt=(Xt 1, . . . , Xt n t ) has a time-varying random number of members, where nt is the number of members at time t. Since the order of members within this collection is immaterial to the problem, we use the empirical measure of the members χt{dot over (=)}Σi=1 n 1 δxi i to represent the household.
  • The household regime represents a current viewing “mindset” of the household that can materially influence the generation of click stream data. The household's current regime rt is a value from the state space R. In one embodiment of the invention, the regimes can consist of values such as “normal,” “channel flipping,” “status checking,” and “favorite surfing.”
  • Thus, the complete signal is composed of the household and the regime:

  • χt=(χt ,R t)
  • which evolves in some state space E.
  • The state of the signal evolves over time via rate functions λ, which probabilistically govern the changes in signal state. The probability that the state changes from state i to j later than some time t is then:

  • R i→j T(t)=P(T>t)=exp(−∫o tλT(s)ds)
  • There are separate rate functions for the evolution of each individual, the household membership itself, and the household's regime. In one embodiment of the invention, the rate functions for an individual i depend only on the given individual, the empirical measure of the signal, the current time, and some external environmental variables λ(t, χt i, χtεt).
  • The number of individuals within the household nt varies over time via birth and death rates. Birth and death rates do not merely indicate new beings being born or existing beings dying—they can represent events that cause one or more individuals to enter and exit the household. These rates are calculated based on the current state of all individuals within the household. For example, in one embodiment of the invention a rate function describing the likelihood of a bachelor to have either a roommate or spouse enter the household may be calculated.
  • In one embodiment of the invention, these rate functions can be formulated as mathematical equations with parameters empirically determined by matching the estimated probability and expected value of state changes from available demographic, macroeconomic, and viewing behavior data. In another embodiment, age can be evolved deterministically in a continuous state space such as [0, 120].
  • 2.2.2 Observation Model Description
  • In general, the observation model describes the random evolution of the click stream information that is generated by one or more individuals' interaction with a DSTB. In one preferred embodiment of the invention, only current and past channel change information is represented in the observation model. Given a universe of M channels, we have a channel change queue at time tk of Yk=(yk, . . . , yk−B+1) with B representing the number of retained channel changes, channels that were watched in the past B discrete time steps. In one preferred embodiment of the invention, only the times when a channel change occurs as well as the channel that was changed to are recorded to reduce overhead.
  • In the more general case, a viewing queue contains this current and past channels as well as such things as volume history. In the aforementioned case, the viewing queue degenerates to the channel change queue.
  • The probability of the viewing queue changing from state i to state j at time t based on the state of the signal and some downloadable content Dt (denoted as pi→j(Dt, Xt)) is then determined. In one preferred embodiment, this downloadable content contains, among other things, some program information detailing a qualitative category description of the shows that are currently available, for instance, for each show, whether the show is an “Action Movie” or a “Sitcom”, as well as the duration of the show, the start time of the show, the channel the show is being played on, etc.
  • In the absence of a special regime, an empirical method has been created to calculate the Markov chain transition probabilities. These probabilities are dependent on the current state of all members of the household and the available programs. This method is validated using observed watching behavior and Varadarajan's law of large numbers. Suppose that P is a discrete probability measure, assigning probabilities to Ω={ω1, . . . , ωK} and we have N independent copies of the experiment of selecting an element. Then, the law of large numbers says that
  • 1 N i = 1 N k = 1 K 1 ω k = ω i P ,
  • where ωi is the ith random outcome of drawing an element from Ω.
  • In one embodiment of the invention, this method focuses on calculating the probabilities for a channel queue of size 1 (i.e., Yk=yk). The observation probabilities, that is, the probabilities of switching between two viewing queues over the next discrete step, can be first calculated by determining the probability of switching categories of the programs and then finding the probability of switching into a particular channel within that category. The first step is to calculate, often in a offline manner, the relative proportion of category changes that occur due to channel changes and/or changes in programs on the same channel. In order to perform this calculation, the set of all possible member states Xt is mapped into a discrete state space Π such that ƒ(Xt)=πt for some πt ε Π for all possible Xt. We suppose there are a fixed, finite set of categories C={c1, c2, . . . , cK}. Furthermore, let there be Nv viewer records, with each viewer record representing a constant period of time Δt, and with each three-tuple viewing record V(k)=(π,b,c) with k=1, 2, . . . , Nσand b, c ε C, containing information about the discretized state of the household (r) and the category at the beginning (b) and the end (c) of the time period. Then, for each π ε II and b, c ε C, we calculate:
  • N ( π , b , c ) = { k = 1 N υ 1 v ( k ) ( π , b , c ) , 0 ,
  • b→c valid this time step,
    otherwise.
  • When the optimal estimation system is running in real-time, the probabilities for the category transition from ci to cj that occurs at a given time step are calculated first by calculating the probability of category changes given the currently available programs:
  • P c i c J ( π ) = [ N ( π , c i , c j ) cx = 1 K N ( π , c i , c α )
  • where the summation from α=1 to K accounts for all of the categories in C. Suppose that ci is the category associated with channel i and cj is the category associated with channel j. Then, this probability is converted into the needed channel transition probability by:
  • P i j ( π ) = P c 1 c J ( π ) n t ( c j )
  • Where nt(cj) is the number of channels that have shows that fall in category c1 at the end of the current time step.
  • An alternative probability measure may be calculated by the “popularity” of channels instead of the transition between channels at each discrete time step. This above method can be used to provide this form by simply summing over the transition probabilities for a given category:
  • P c J ( π ) = α = 1 K N ( π , c α , c j ) β , γ = 1 K N ( π , c β , c γ ) .
  • Again, this probability is converted into the needed channel transition probability by using an instance of multiplication rule:
  • P j ( π ) = P c J ( π ) n t ( c j ) ,
  • Where, again, nt(cj) is the number of channels that have shows that fall into category cj at the end of the current time step.
  • In one embodiment of the invention, several or all of the categories will be programs themselves, given the finest level of granularity. In other instances, it is preferable to have broad categories to reduce the number of probabilities that need to be stored down.
  • 2.3 Optimal Estimation with Markov Chain Observations
  • In the traditional filtering theory summarized above, one has that the observations are a distorted, corrupted partial measurement of the signal, according to a formula like

  • Y k =ht k ,V k),
  • where tk is the observation time for the kth observation and {VK}k=1 00 is some driving noise process, or some continuous time variant. However, for the DSTB model that we described in the immediately previous subsections, we have that Y is a discrete time Markov chain whose transition probabilities depend upon the signal. In this case, the new state Yk can depend upon its previous state, rendering the standard theory discussed above invalid. In this section, a new, analogous theory and system is presented for solving problems where the observations are a Markov chain. One noticeable generality of the system is that Markov chain observations may only be allowed to transition to a subset of all the states, a subset that depends on the state that the chain is currently in. This is a useful feature in the targeted advertising application, since much of the viewing queue's previous data may remain in the viewing queue after an observation and the insertion of some new data. For assimilation ease, this is described in the context of targeted advertisement even though it clearly applies in general.
  • Suppose that we have a Markov signal Xt with generator
    Figure US20130254787A1-20130926-P00001
    and with an initial distribution v. Recall that the signal Xt evolves within the state space E. To be precise, the signal is defined to be the unique DE{0,∞) process that satisfies the (
    Figure US20130254787A1-20130926-P00001
    , v)-martingale problem:

  • P(X 0 ,E,)=σ()

  • and

  • M t(φ){dot over (=)}φ(X t)−φ(X 0)−∫0 t
    Figure US20130254787A1-20130926-P00001
    φ(X x)ds
  • is a martingale for all φ ε D(
    Figure US20130254787A1-20130926-P00001
    ).
  • We wish to estimate the conditional distribution of Xt based upon {1, 2, . . . , M}-valued discrete-time Markov chain observations that depends upon Xt as well as some exogenous information Dt. Recall that Yk=(yk, . . . yk−B+1), with B representing the number of retained channel changes. To make things manifest, suppose that {vk}k=−∞ is a sequence of independent random variables that are independent of the signal and observation such that
  • P ( v k = i ) = 1 M
  • for i=1, 2, . . . , M and k ε Z and that the observation y k occurs at time tk with finite state space {1, . . . , M} of events available, where
  • y k = v k k = 0 , - 1 , - 2 , y k k = 1 , 2 , 3 ,
  • transitions between values in {1, . . . , M} with homogeneous transition probabilities pi→j(Dt1Xi) of going from state i to state j at time t. Here, Dt and Xt are the current states of the pertinent exogenous information and signal states at the time of the possible state change.
  • To ease notation, we define Dk{dot over (=)}Dt tXk{dot over (=)}Xt k and set
  • Vk = ( u k 1 u k - 1 , uk - B + 1 ) T for k = 1 , 2 , Z j = { k = 1 j Ϛ k - 1 ( X k ) for j = 1 , 2 , 1 for j = - 1 , - 2 , and Z t Z j for t ( t j , t j + 1 ) , where Ϛ k ( X k ) = M × p γ k - 1 γ k ( D k , X k ) .
  • Then, some mathematical calculations show that
  • E [ f ( X t ) | σ { Y 1 , , Y j } ] = E _ f ( X t ) ( Z ( T ) ) - 1 | σ { Y 1 , , Y j } E _ [ ( Z ( T ) ) - 1 | σ { Y 1 , , Y j } ] , for t j T , where f : E R and P _ ( A ) = E [ 1 A Z ( T ) ] A σ { ( X t , Y t ) , t T } . Letting η ( t ) 1 Z ( t ) , ( 1 )
  • and noting the denominator and numerator of equation (1) above are both calculated from Ē[g(Xt)η(t)|]Ft γ.
    with g=1 and g=ƒ respectively, where

  • F t γ
    Figure US20130254787A1-20130926-P00002
    σ{Y 1 , . . . , Y j} for t ε[t j ,t j+1),
  • we just need an equation for

  • μt ƒ{dot over (=)}Ē[f(X t)η(t)|F t γ]
  • for a rich enough class of functions ƒ:E→R.
  • More mathematics establishes that μt(dx)
    Figure US20130254787A1-20130926-P00002
    Ē(1x t εdx(t)|Ft γ) satisfies
  • μ t ( ϕ ) - μ 0 ( ϕ ) = 0 t μ s ( ) s + k = 1 n t μ t k ( ϕ ζ _ k )
  • for all t ε[0,∞) and φ ε D(
    Figure US20130254787A1-20130926-P00001
    ), where
  • ζ _ k ( x ) = 1 - 1 ζ k ( x )
  • and ns=max {k:tk≦s}.
  • 2.4 Filtering Approximations
  • In order to use the above derivation in a real-time computer system, approximations must be made so that the resulting equations can be implemented on the computer architecture. Different approximations must be made in order to use a particle filter or a discrete space filter. These approximations are highlighted in the sections below.
  • 2.4.1 Particle Filter Approximation
  • By equation (1) we only need to approximate
  • μ t ( ds ) E _ [ 1 X t dx η ( t ) | F t γ ] , where η ( t ) = k = 1 t M × p γ k - 1 γ k ( D k , X k ) = k = 1 t M × p γ k - 1 γ k ( D k , X t k )
  • is the weighting function. Now, suppose that we introduce signal particles {ξt i,t≦0}i=1 , which evolve independently of each other, each with the same law as the historical signal, and define the weights
  • η i ( t ) = k = 1 t M × p γ k - 1 γ k ( D k , ξ t k i ) ,
  • Then, it follows by deFinnetti's theorem and the law of large numbers that
  • 1 N i = 1 N η i ( t ) δ ξ t i ( dx ) μ t ( dx ) .
  • 2.4.2 Discrete Space Approximation
  • If we can assume that the state space of E of Xt is a compact metric space, then for each N ε N, we let lN and MN satisfy lN→∞ and MN→∞ as M→∞. For DN={1, . . . dN}⊂ N, we suppose that {Ck N, k ε DN} is a partition of E such that
  • max k diam ( C k N ) N 0 ,
  • and for large enough N that all the discrete state components are in different cells. Then, we take yk N ε Ck N and define JN{dot over (=)}{0, 1, . . . , MN}d N . Take η(CN)=j to mean η(Ci N)=ji for all i ε DN and η ε Mc f(E). Then, the unnormalized distribution of the signal μt u satisfies
  • μ t ( η ( C N ) = j ) = μ 0 ( η ( C N ) = j ) + 0 t μ s ( N 1 η ( C N ) = j ) s + k = 1 n t μ t k ( 1 { η ( C N ) = j } ζ _ k )
  • where
    Figure US20130254787A1-20130926-P00001
    N is some discretized version of
    Figure US20130254787A1-20130926-P00001
    . The application of REST then creates particle counts {Nt c,p} for each cell in CN and for each household population p within the cell-dependent set of allowable populations Pc N, such that
  • μ t N ( dx ) = c C N p P c N n t c , p δ p , c ( dx ) .
  • Then, it follows that

  • μt N(dx)
    Figure US20130254787A1-20130926-P00003
    μt(dx)
  • as N→∞ for each t≧0.
  • 2.5 Refining Stochastic Grid Filter with Discrete Finite State Spaces
  • In U.S. Pat. No. 7,188,048, a general form of the REST filter was detailed. This method and system has demonstrated to be of use in several applications, particularly in Euclidean space tracking problems as well as discrete counting measure problems. However, several improvements upon this method have been discovered, which provide dramatic reductions in the memory and computational requirements for an embodiment of the invention. A new method and system for the REST filter is described herein where the signal can be modeled with a discrete and finite state space. Examples using the targeted advertising model are provided for clarity, but this method can be used with any problem that features the environment discussed below.
  • 2.5.1 Environment Description
  • In certain problems, the signal is composed of zero or more targets Xt i and zero or more regimes Rt j. For example, in targeted advertising one embodiment of the signal model is in the form χt=(Xt, Rt), where xt is the empirical measure of the targets (or, more specifically, the household members) and there is only one regime. Furthermore, each target and regime have only a discrete and finite number of states, and there are a finite number of targets and regimes (and consequently a finite number of possible combinations of targets and regimes). The finite number of combinations need not be all possible combinations—only a finite number of legitimate combinations are required. For instance, a finite number of possible types of households (meaning households that exhibit particular demographic compositions within) can be derived from geography-dependent census information at relatively granular levels. Instead of having all potential combinations of individuals (up to some maximum household membership nMAX), only those combinations which can be possibly found within a given geographic region need to be considered legitimate and contained within the state space.
  • In these restricted problems, some components of the state of the target(s) and/or regime(s) may be invariant over the short period during which the optimal estimation is occurring. In these cases, such state information is held to be constant, while other portions of the state information remain variant. In one embodiment of the household signal model, the age, gender, income, and education levels of each individual within the household may be considered to be constant, as these values change over longer periods of time and the DSTB estimation occurs over a period of a few weeks. However, the current watching status and household regime information will change over relatively short time frames, and as a result these states are left to vary in the estimation problem. We shall denote the invariant portion of the signal as {circumflex over (X)} and the variant portion of the signal as {tilde over (X)}. There are N possible invariant states (the ith such state donated by {circumflex over (X)}i) and Mi possible variant states for the ith invariant state (the jth state denoted by {circumflex over (X)}i,j).
  • 2.5.2 REST Finite State Space System Overview
  • FIG. 29 depicts one preferred embodiment of the REST filter in a finite state space environment. REST is composed of a collection of invariant state cells, each of which represents one possible collection of targets and regimes for the signal along with their invariant state properties. Each invariant cell contains a collection of variant state cells, each representing the possible time-variant states of the given invariant cell. Implicitly, the variant cells contain the invariant state information of their parent invariant cell, meaning each variant cell represents a particular potential state of the signal. The invariant cells themselves represent an aggregate container object only and are used for convenience purposes. The collections of variant and invariant cells may be stored on a computer medium in the form of arrays, vectors, list or queues. Cells which have no particle count at a given time t may be removed from such containers to reduce space and computational requirements, although a mechanism to reinsert such cells at a later date is then necessary.
  • As shown in FIG. 30, each variant state sell contains a particle count nt i,j. This particle count represents the discretized amplitude of that cell. As noted previously, this amplitude is used to calculate the conditional probability of a given state. Each variant state cell also contains a set of imaginary clocks λt i,j,q. These imaginary clocks represent the time varying progression towards the event of a particle count change within a cell driven by both continuous transition rates and discrete observation events. For each variant state cell there are Qi,j possible state transitions. In this environment, all valid state transitions occur within the same invariant state cell. To account for simultaneous changes in the conditional distribution of the REST filter, a temporary particle counter entitled particle count Δnt i,j is used to store the number of particles that will be added or removed from the given variant state cell once the sequential processing of all cells is completed. Cells which have a valid state transition from the variant state cell with state {tilde over (X)}i,j are said to be neighbors of that cell.
  • As mentioned above, the invariant state cells are containers used to simplify the processing of information. Each invariant state cell's particle count nt i is an aggregate of its child variant state cell particle counts. Similarly, the invariant state cell's imaginary time clock is an aggregation of all clocks from the variant cells. This aggregation facilitates the filter's evolution, as invariant states which have no current particle count can be skipped at various stages of processing.
  • 2.5.3 REST Filter Evolution
  • FIG. 31 depicts the typical evolution of the REST filter. This evolution method updates the conditional distribution of the filter over some time period Δt by transferring particles between neighboring cells using the imaginary clock values. The movement of a particle between neighboring cells is known as an event. (In practice, the movement of particles can be replaced with equivalent births and deaths to allow efficient cancellation of opposite rates.) Such events are simulated en masse to reduce the computational overhead of the evolution. The number of events to simulate is based on the total imaginary clock sum λt for all cells. FIG. 32 shows the method that determines how particles move to each neighboring cell. When the simulation of events is complete, the particle counts are updated and the imaginary clocks are scaled back to represent the change in the state of the filter.
  • Compared to the previous method described in U.S. Pat. No. 7,188,048, additional steps have been added to improve the effectiveness of the filter. Specifically, an adjustment to the cell particle counts now occurs prior to the push down observations method, and a drift back routine has been added prior to particle control. In certain problems, some cell states may have no possibility of being the current signal state based on observation information. For instance, a household must have a least one member currently watching if a channel change is recorded. In these circumstances, the particles in all invalid states must be redistributed proportionately to valid states. Thus, if there are nt invalid particles to redistribute, then all valid variant state cells will receive
  • n t invalid n t i , j i , j n t i , j
  • particles, and will receive an additional particle with probability
  • n t invalid n t i , j i , j n t i , j - n t invalid n t i , j i , j n t i , j .
  • When this type of observation-based adjustment is used, it is likely that the rates governing the evolution of the signal must be appropriately altered to coincide with the use of observation data in this manner.
  • To improve the robustness of the REST filter, a drift back method has been added. This method uses some function ƒ({tilde over (X)}i,j,t) to add nt seed particles to variant state cells based on the initial distribution v of the signal. The number of particles to add to each cell depends on time, the given cell, and the overall state of the filter. This method ensures that the filter does not converge to a small set of incorrect states without the ability to recover from an incorrect localization.
  • 2.6 Head End Estimations
  • In order to maximize the profitability of multiple service operators' advertising operations, the determination of which commercials to distribute to a collection of DSTBs is critical. As more information is available about the actual viewership of commercials based on the conditional distributions (or conditional estimates derived thereof) of a DSTB-based asymptotically optimal nonlinear filter, the pricing of specific commercial slots can be more dynamic, thus improving overall profits.
  • To capitalize upon this potential, an estimate of the collection of household probability distributions, that includes such things as the number of people within each demographic, is performed at the Head End based on the whole set or a random sampling of conditional DSTB estimates. The following model contains a prefer embodiment of the Head End estimation system.
  • 2.6.1 Head End Signal Model
  • The Head End signal model consists of pertinent trait information of potential and current television viewers that have DSTB, in communication with a particular Head End. A state space S is defined that represents such a collection of traits for a single individual. In one embodiment of the invention, this space could be made up of age ranges, gender, and recent viewing history for an individual. To keep track of individuals, we let C0=0 be the household type of no individuals and Cn be the collection of household types with n individuals

  • C n={((s 1 n 1), . . . , (s r ,n r)):s i εS and distinct, n 1 +n 2 + . . . +n r +n}.
  • The collection of households would then be the union
  • n = 0 C n
  • of the households with n people in them. Realistically, there would be a largest household N that we could handle and we set the household state space to be
  • E = n = 0 N C n ,
  • where N is some large number.
  • To process the estimate transferred back from the DSTBs through the random sample mechanism, we also want to track the current channel for each DSTB. This means that each DSTB state; including potential household viewership, watching status, and current channel; is taken from

  • D{dot over (=)}E×{1, 2, . . . , M},
  • where there is M possible channels that the DSTB could be tuned to.
  • We are not worried about a single DSTB nor even which DSTBs are in a particular state but rather with how many DSTBs are in state d ε D. Therefore, we let X to be tracked, be a finite counting measure valued process, counting the number of DSTBs in each category d ε D over time. For technical reasons we define the signal to be either the probability distribution of X of the probability distributions of each component of X.
  • In an embodiment of the invention, it is possible to track in aggregate the possible number of DSTBs in each category to minimize the computational requirements. In such a case, elements of size o are used so that the total will still sum to the maximum number of DSTBs. For example, suppose that there are 1 million DSTBs. Then, we would have 100,000 elements (consisting of a=10 DSTBs each) distributed over D. Suppose M(D) denotes the counting measure on D and M(D) denotes the subset of M(D) that has exactly 100,000 elements. The signal will evolve mathematically according to a martingale problem

  • ƒ(X t)=ƒ(X 0)+∫0 t
    Figure US20130254787A1-20130926-P00001
    ƒ(X s)ds+M t(ƒ),
  • where t→Mt(ƒ) is a martingale for each continuous, bounded functional ƒ on M(D) and
    Figure US20130254787A1-20130926-P00001
    is some operator that would be determined largely from the DSTB rates and the natural assumption that the households act independently.
  • Any households that provide their demographics in exposed mode are not considered to be part of the signal.
  • 2.6.2 Head End Observation Models Herein we describe two observation models: one for the random sampling of DSTBs and one for delivery statistics.
  • For the random sample observation model, we consider the channel and viewership by letting X be our process as in the previous section, and let Vk denote the random selection at time tk in the sampling process. To be precise, suppose that there are M DSTBs for a particular Head End and suppose that a DSTB that believes at least one person is currently watching will supply a sample with a fixed probability of five percent. Then, Vk would be a matrix with a random number of rows, each row consisting of M entries with exactly one nonzero entry corresponding to the index of the particular DSTB which has provided a sample. The number rows would be the number of DSTBs providing a sample. The locations of the nonzero entries are naturally distinct over the rows and would be chosen uniformly over the possible permutations to reflect the actual sampling taken.
  • Now, we let ({circumflex over (P)}t k , Uk) be the (column) vectors of the conditional distribution viewership estimates and corresponding channel changes of the M DSTBs, all at time tk. Then, this observation process would be

  • θt k 1 =h(V k({circumflex over (P)} k , U k)).
  • Here, the Vk would do the random selection and the h would be a function providing the information that is chosen to be communicated to the Head End.
  • For the aggregated ad delivery statistics model, we have time-indexed sequences of functions Hk,j that provide a count of the various ads delivered previously at time tk−tj. There would be a small amount of noise Wk,j due to the fact that some DSTBs may not return any information due to temporary malfunction (i.e. a ‘missed observation’), and due to the fact that the estimated viewership used to determine a successful delivery is not guaranteed to be correct.
  • The second observation information from the aggregated delivery statistics would be

  • θt k 2,j =H k,j({circumflex over (P)} t k −t j ,W k,j).
  • Here, j ranges back over the spot segments in the reporting periods and tk is the reporting period time.
  • 2.6.3 Head End Filter
  • In a preferred embodiment of the invention, the signal for the Head End is taken to be a representation for the probability distributions from the DSTBs. This assignment can make the estimation problem more workable.
  • 2.7 Head End Commercial Selection
  • In certain embodiments of the invention, other information may be available which also can be used to perform the aggregate viewership estimation. For example, aggregate (and possibly delayed) ad delivery statistics can also provide inferences in the estimated viewership of DSTBs, as well as any ‘exposed mode’ information whereby households opt to provide their state information (demographics, psychographics, etc.) in exchange for some compensation.
  • In this setting, commercial contract is modeled as a graph of incremental profit in terms of the contract details, available resources and future signal state. We call these graphs contract graphs which arrive with rates that depend upon the contract details, signal state and economic environments. Some of the contract details may include:
  • Number of times commercial is to be shown (could contain minimum and maximum thresholds), likely in thousands;
  • Time range for time of day/week that commercial is to be shown;
  • The Target demographic(s) for the commercial;
  • Particular channels or programs that the commercial is to be shown on; and
  • Customer that wrote the contract.
  • The random arrival of the contract graphs is denoted as the contract graph process. Furthermore, an allotment of resources (that need not be the maximum allotable to any contract) to a contract graph process is called a feasible selection if, given the state (present and future) and the environment, the allotted resources do not exceed the available resources, i.e. the available commercial spots over the various categories. Now, due to the fact that these limited resource become depleted as one accepts contracts, current versus future potential profits are modeled through a utility function. This utility function takes the stream of contract graphs available (both presently and with future random arrivals) and returns a number indicating profit in terms of dollars or some other form of satisfaction. Due to the random future behavior of contract graphs, the utility function cannot simply provide maximum profits without taking into account deviation from the expected profit to ensure the maximization does not allow significant risk of poor profit.
  • To perform optimal commercial selection, the following models need to be defined: the Head End signal model, the Head End observation model, the contract generation model, and the utility (profit) model.
  • 2.7.1 Contract Model
  • The commercial contracts that arise are modeled as a marked point process over the contract graphs. The rate of arrival for the contracts depends upon the previous contracts executed as well as external factors such as economic conditions.
  • Suppose that 1 denotes Lesbegue measure. Then, we let C denote the space of possible contract graphs with some topology on it, {ηt, t≧0} denote the counting measure stochastic process for the arrival of contract graphs up until time t and ξ denote a Poisson measure over C×[0,∞)×[0,∞) with some mean measure v×l×l. Furthermore, we let λ(c,η[0,t), t) be the rate (with respect to v) that a new contract will come with contract graph c ε C at time t when η0,t) the records the arrival of contract graphs from time 0 up to but not including time t. Then, we model contract arrival by the following stochastic differential equation”

  • n t(A)=η0(A)+∫A×[0,∞)×[0,t]1[0,λ(c,η o,s ,s))(v)ξ(dc×dv×ds) for all AεB(C).
  • It is possible that the contract details noted above may be altered upon acceptance of a contract. As a result, the contract details are modeled to depend on an external environment which can evolve over time.
  • 2.7.2 Utility Function Description
  • To ease notation, we let R(DS) be the available resources, now and in the future, based upon the downloadable program information DS at time s.
  • We will not be able to accept all contracts that arise and we have to make the decision whether to accept or reject a contract without looking into the future. We denote an admissible selection as a feasible selection such that each resource allocation decision does not use future contract or future observation information. In terms of the notation of the previous section, we suppose that nt represents the number of contracts that have arrived of the various types up to and including time t and take
  • γ t ( l ) = Q C × [ 0 , t ] c ( l s - , X s - , q ) η ( c × s ) q for each t 0 ,
  • where Q represents the set of all potential customers and {ls,s≧0} is a selection process, i.e., allocates resources to each contract c. Then, {s≧0} is an admissible selection if ls≦R(Ds) for each s≧0 and ls does not use future contract or observation information, i.e., is measurable with respect to σ({ηu,u≦s}, {θt k 1t k 2,j j ε N, tk≦s}) for each s≧0. Now, γt(l) represents the profit obtained up to time t through admissible selection l. To ease notation, we let Λ be the set of all such admissible selections.
  • The utility function J balances current profit with future profit and the chance of obtaining very high profits on a particular contract with the risk of no or low profit. In order to ensure that we start off reasonably, we will deweight future profit in an exponential manner. Moreover, in order that we are not overly aggressive we will include a variance-like condition. One embodiment of the resulting utility function is

  • J(X,l)=∫[0,∞) e −λtt(l)−α(γt(l))2 ]dt,
  • for small constants λ, a>0. Then, the goal of the commercial selection process is to maximize E[J(X,l)] over the l ε Λ. Such a goal can be solved using one or more asymptotically optimal filters.
  • In summary, a matching algorithm may involve the following steps. First, an ADR is received from matching to one or more of the identified viewers. The similarity function is applied with respect to each parameter, and the results are aggregated as discussed above in order to determine a degree of match. Once the classifier has discovered one or more viewers that match the basic targeting parameters of the ad, the classifier determines whether or not any of these viewers are currently watching the television set. To do this, the periodicity analysis function is called to match a week by day and week by time of day terrain surface to the required time. This process returns a value reflecting the expected degree of match between the viewing time behavior of each viewer and the time frame of the ad. The degree of match in this regard is used to scale the base compatibility score to produce a time compatibility score. Once the classifier has thus identified a set of compatible viewers and determined which of these viewers are currently watching the television (that is, the time compatibility value is greater than some threshold), the classifier determines whether or not to accept this ad (vote for the ad or select the ad for delivery) based on any frequency limitation constraints and also subject to consideration of ad importance. That is, there may be additional constraints associated with the ad regarding the frequency with which the ad may be delivered to an individual viewer or the total number of times that the ad may be delivered to a viewer. The frequency analyzer returns a value and range of zero to one that is used to scale the time compatibility rank. This creates a frequency compatible rank. In the case of ad importance, a first ad may be selected (and voted) rather than a second ad, where the ads have a similar degree of “match” or even where the second ad has a better match, due to differences in ad importance (e.g., where the first ad has a higher importance).
  • An ad may also have constraints. For example, the ad may have target age limitations, genre restrictions, network restrictions, program nature restrictions, rating restrictions or the like. The placement constraint analyzer that meets the above-noted compatibility requirements and then searches for any required placement constraints. The placement analysis returns a gateway value of one or zero where one indicates no active placement constraints and zero indicates placement constraints that are violated. This creates the final compatibility index score.
  • It will be appreciated that the matching process need not be based on a continuous, or even finely graded, value range. For example, the result of the matching process may be a binary “match” or “no match” determination. In this regard, a threshold, or set of thresholds with associated decision logic, may be used to define a match or lack thereof for an ad with respect to a current audience.
  • As shown in FIG. 10, the classifier may thus be viewed as incorporating a number of functional components including a stay transition and link manager, a feature acquisition subsystem, a mountain clustering subsystem, periodicity analysis subsystem and an ad matching and ranking subsystem. The stay transition and link manager are operative to monitor the click stream and determine whether the DSTB is turned on as well as tracking stay changes. The feature acquisition subsystem is operative to build the initial feature terrain as discussed above. The mountain clustering subsystem is operative to process the feature terrain to remove noise and better define viewer sites as discussed above. The periodicity analysis subsystem recognizes viewing patterns and matches those patterns to viewers as identified from the processed feature terrain space. Finally, the ad matching and ranking subsystem compares viewer classification parameters to targeting parameters of an ad and also analyzes viewing habits of the viewer in relation to the delivery time, so as to match ads to viewers and develop a ranking system for voting, ad selection and reporting.
  • The foregoing description of the present invention has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit the invention to the form disclosed herein. Consequently, variations and modifications commensurate with the above teachings, and skill and knowledge of the relevant art, are within the scope of the present invention. For example, although fuzzy sets and fuzzy rules, stochastic filtering, etc. have been described in connection with various processes above, at least some aspects of the present invention can be implemented without such features. The embodiments described hereinabove are further intended to explain best modes known of practicing the invention and to enable others skilled in the art to utilize the invention in such, or other embodiments and with various modifications required by the particular application(s) or use(s) of the present invention. It is intended that the appended claims be construed to include alternative embodiments to the extent permitted by the prior art.

Claims (49)

1. A method for use in targeting assets to users of user equipment devices in a communications network, comprising the steps of:
operating a processor to progressively incorporate, over time, a plurality of user inputs by one or more users of a user equipment device into a model of a user composition of the one or more users, the model including a plurality of user classification parameters;
filtering the user composition model to obtain an estimate of a current user composition of the user equipment device; and
targeting one or more assets in the communications network using the estimated current user composition.
2. The method of claim 1, wherein the filtered user composition model is free of user equipment device usage patterns for the users obtained before the progressive incorporation of the plurality of user inputs.
3. The method of claim 1, further comprising:
receiving first user inputs at a first time; and
receiving second user inputs at a second time after the first time, and wherein the second user inputs are incorporated into the model to a greater degree than are the first user inputs.
4. The method of claim 1, further comprising:
establishing a reference event, wherein the progressively incorporated inputs occurred at times after the reference event.
5. (canceled)
6. (canceled)
7. The method of claim 1, wherein the targeting comprises:
receiving one or more lists of assets for delivery at the user equipment device;
obtaining one or more targeting parameters for the one or more lists of assets; and
determining a level of correspondence between the user classification parameters of the current user composition of the user equipment device and the targeting parameters for the one or more lists of assets.
8. The method of claim 7, wherein the determining is performed in multiple dimensions relating to multiple classification and targeting parameters.
9. The method of claim 8, further comprising:
voting for at least one asset of the one or more lists of assets based on the determined level of correspondence between the user classification parameters of the current user composition of the user equipment device and the targeting parameters for the one or more lists of assets.
10-13. (canceled)
14. The method of claim 1, further comprising:
progressively incorporating, over time, a plurality of user inputs by users of a plurality of additional user equipment devices into a plurality of models of user composition of the users, the models including a plurality of user classification parameters;
filtering the user composition models to obtain estimates of current user compositions of the plurality of additional user equipment devices; and
aggregating the current user compositions of the user equipment device and the plurality of additional user equipment devices to obtain a current user composition of an aggregated audience, wherein the targeting comprises using the aggregated audience current user composition for use in targeting one or more assets in the communications network.
15. (canceled)
16. The method of claim 1, wherein the progressively incorporating comprises:
associating the user inputs with respective user classification parameters;
treating the values as points in fuzzy sets; and
aggregating the fuzzy sets into a multi-dimensional feature terrain.
17. The method of claim 16, wherein the filtering comprises:
reducing noise in the feature terrain.
18. The method of claim 17, wherein the reducing comprises:
processing evidence in relation to reference values.
19. The method of claim 17, wherein the reducing comprises:
reducing a set of gradients across the feature terrain by iteratively removing low-level gradients to progressively identify candidate centers.
20. The method of claim 1, wherein the progressively incorporating comprises:
developing an observation model based on first inputs by one or more users with respect to one or more user equipment devices; and
developing a signal model reflective of the possible states and dynamics of a user composition of one or more users of a first user equipment device with respect to time, wherein said observation model probabilistically relates measurement data related to said first inputs to the possible states and dynamics.
21. The method of claim 20, wherein the filtering the user composition model comprises:
employing a stochastic filter to estimate said user composition at a time of interest through an approximate conditional distribution of a signal given the signal and observation models and second inputs by one or more users.
22. The method as set forth in claim 21, wherein said inputs are a click stream of user inputs over time and said observation model models said click stream as a Markov chain.
23. The method as set forth in claim 22, wherein said observation model takes into account programming related information for network content indicated by at least some of said inputs.
24. The method as set forth in claim 23, further comprising the step of processing said Markov chain using a mathematical model wherein observations of said Markov chain may only transition to a subset of a full set of states, where said subset depends on a current state of said Markov chain.
25. The method as set forth in claim 21, wherein said step of developing an observation model comprises modeling said observation model as a Markov chain or a k step Markov chain.
26. The method as set forth in claim 25, wherein the transition function for the observation Markov chain depends upon a position of the signal to estimate.
27. The method as set forth in claim 21, wherein said signal is established as representing said user composition and a separate factor affecting said user inputs.
28. The method as set forth in claim 21, wherein a model of said signal allows for representation of said user composition as including two or more users.
29. The method as set forth in claim 21, wherein a model of said signal allows for representation of a change in said user composition.
30. The method as set forth in claim 29, wherein said change is a change in a number of users associated with said user equipment device.
31. The method as set forth in claim 21, wherein said step of employing a stochastic filter comprises obtaining probabilistic estimates of said signal based on said observation model and measurement data.
32. The method as set forth in claim 31, wherein said step of employing a stochastic filter comprises defining a nonlinear filter to obtain probabilistic estimates of said signal based on said observation model and measurement data.
33. The method as set forth in claim 32, wherein said step of employing a stochastic filter further comprises establishing an approximation filter for approximating operation of said nonlinear filter.
34. The method as set forth in claim 33, wherein said approximation filter is a particle filter.
35. The method as set forth in claim 33, wherein said approximation filter is a discrete space filter.
36. The method as set forth in claim 21, wherein said step of using comprises providing information based on said user composition to a network platform operative to insert assets into a content stream of said network.
37. The method as set forth in claim 36, wherein said information identifies demographics of one or more users of said user equipment device.
38. The method as set forth in claim 37, wherein said platform is operative to aggregate user composition information associated with multiple user equipment devices and to select one or more assets for insertion based on said aggregated information.
39. The method as set forth in claim 38, wherein said platform is operative to process information from multiple user equipment devices as an observation model and to apply a filter with respect to said observation model to estimate an aggregate composition of a network audience at said time of interest.
40. The method as set forth in claim 37, wherein said platform is operative to select assets for insertion based on said aggregate composition and additional information affecting a delivery value of particular assets.
41. The method as set forth in claim 36, wherein said information identifies one or more appropriate assets for delivery to said user equipment device based on said user composition.
42. The method as set forth in claim 21, wherein said step of using comprises selecting, at said user equipment device, an asset for delivery to said one or more users.
43. The method as set forth in claim 21, wherein said step of using comprises reporting a goodness of fit of an asset delivered at said user equipment device with respect to said one or more users.
44. An apparatus for use in targeting assets to users of user equipment devices in a communications network, comprising:
a port operative for receiving input information regarding first inputs by one or more users with respect to a user equipment device; and
a processor operative for providing an observation model based on said first inputs, modeling the observation model as dependent upon a signal model reflective of at least a user composition of one or more users of said user equipment device with respect to time, where said observation model probabilistically relates measurement data related to said first inputs to said user composition, employing a stochastic filter to estimate the user composition at a time of interest as a state of a signal through an approximate conditional distribution of the signal given the signal and observation models and second inputs by one or more users, and using the estimated user composition in targeting an asset with respect to the user equipment device.
45. The apparatus as set forth in claim 44, wherein said processor is operative for defining a nonlinear filter to obtain estimates of said signal based on said observation model and measurement data.
46. The apparatus as set forth in claim 45, wherein said processor is operative for establishing an approximation filter for approximating operation of said nonlinear filter.
47. The apparatus as set forth in claim 46, wherein said nonlinear filter is one of a particle filter and a discrete space filter.
48. The apparatus as set forth in claim 44, further comprising a port for transmitting information for use in targeting assets to a separate network platform, wherein said information is based on said estimated user composition.
49. A method for use in targeting assets to users of user equipment devices in a broadcast network, comprising the steps of:
collectively analyzing a stream of data corresponding to a series of first user inputs with respect to one or more user equipment devices, wherein said step of collectively analyzing comprises establishing an observation model; and
applying logic for matching a pattern described by a stream corresponding to a series of second user inputs to a characteristic associated with an audience classification of a user, wherein said step of applying logic comprises employing a stochastic filter to approximately estimate the conditional distribution of a signal given the observation model and second inputs and extract signal estimates from said series of second user inputs to estimate said audience classification at a time of interest.
50. The method as set forth in claim 49, wherein said series of user inputs are modeled as a Markov chain.
51. The method as set forth in claim 49, wherein said step of applying logic comprises using a nonlinear filter model.
52. The method as set forth in claim 51, wherein said step of applying logic comprises executing an approximation filter to approximate operation of said nonlinear filter.
US13/663,780 2006-05-02 2012-10-30 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising Abandoned US20130254787A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US13/663,780 US20130254787A1 (en) 2006-05-02 2012-10-30 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising
US14/949,442 US9693086B2 (en) 2006-05-02 2015-11-23 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising

Applications Claiming Priority (7)

Application Number Priority Date Filing Date Title
US74624406P 2006-05-02 2006-05-02
US74624506P 2006-05-02 2006-05-02
US11/743,544 US7698236B2 (en) 2006-05-02 2007-05-02 Fuzzy logic based viewer identification for targeted asset delivery system
US11/944,078 US20090133058A1 (en) 2007-11-21 2007-11-21 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising
US12/758,532 US20110067046A1 (en) 2006-05-02 2010-04-12 Fuzzy logic based viewer identification for targeted asset delivery system
US13/447,071 US20120204204A1 (en) 2007-11-21 2012-04-13 Method and Appartus to Perform Real-Time Audience Estimation and Commercial Selection Suitable for Targeted Advertising
US13/663,780 US20130254787A1 (en) 2006-05-02 2012-10-30 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising

Related Parent Applications (2)

Application Number Title Priority Date Filing Date
US12/758,532 Continuation-In-Part US20110067046A1 (en) 2006-05-02 2010-04-12 Fuzzy logic based viewer identification for targeted asset delivery system
US13/447,071 Continuation-In-Part US20120204204A1 (en) 2006-05-02 2012-04-13 Method and Appartus to Perform Real-Time Audience Estimation and Commercial Selection Suitable for Targeted Advertising

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US14/949,442 Continuation US9693086B2 (en) 2006-05-02 2015-11-23 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising

Publications (1)

Publication Number Publication Date
US20130254787A1 true US20130254787A1 (en) 2013-09-26

Family

ID=49213572

Family Applications (2)

Application Number Title Priority Date Filing Date
US13/663,780 Abandoned US20130254787A1 (en) 2006-05-02 2012-10-30 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising
US14/949,442 Expired - Fee Related US9693086B2 (en) 2006-05-02 2015-11-23 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising

Family Applications After (1)

Application Number Title Priority Date Filing Date
US14/949,442 Expired - Fee Related US9693086B2 (en) 2006-05-02 2015-11-23 Method and apparatus to perform real-time audience estimation and commercial selection suitable for targeted advertising

Country Status (1)

Country Link
US (2) US20130254787A1 (en)

Cited By (84)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8910195B1 (en) * 2014-02-20 2014-12-09 Google Inc. Systems and methods for enhancing audience measurement data
US20150120848A1 (en) * 2013-10-30 2015-04-30 Mesh Labs Inc. Method and system for filtering electronic communications
WO2015139119A1 (en) * 2014-03-19 2015-09-24 Verosource Solutions Inc. System and method for validating data source input from a crowd sourcing platform
EP2945113A1 (en) * 2014-05-14 2015-11-18 Cisco Technology, Inc. Audience segmentation using machine-learning
US20160094894A1 (en) * 2014-09-30 2016-03-31 Nbcuniversal Media, Llc Digital content audience matching and targeting system and method
US20160094883A1 (en) * 2014-09-29 2016-03-31 Time Warner Cable Enterprises Llc Apparatus and methods for enabling presence-based and use-based services
US20170006342A1 (en) * 2015-07-02 2017-01-05 The Nielsen Company (Us), Llc Methods and apparatus to correct errors in audience measurements for media accessed using over-the-top devices
US20170127110A1 (en) * 2015-11-02 2017-05-04 Turner Broadcasting System, Inc. Generation of reach, mixture, and pricing utilizing a framework for audence rating estimation
US20170150331A1 (en) * 2013-03-16 2017-05-25 Wellpepper, Inc. Adaptive notification system for mobile devices
US20170195714A1 (en) * 2016-01-05 2017-07-06 Gracenote, Inc. Computing System with Channel-Change-Based Trigger Feature
US20170208353A1 (en) * 2016-01-20 2017-07-20 Samsung Electronics Co., Ltd. Content selection for networked media devices
US20170220670A1 (en) * 2016-02-03 2017-08-03 Mx Technologies, Inc. Automated data supplementation and verification
US9788149B2 (en) 2012-08-30 2017-10-10 Time Warner Cable Enterprises Llc Apparatus and methods for enabling location-based services within a premises
US9838754B2 (en) 2015-09-01 2017-12-05 The Nielsen Company (Us), Llc On-site measurement of over the top media
US9872072B2 (en) 2016-03-21 2018-01-16 Google Llc Systems and methods for identifying non-canonical sessions
US9918345B2 (en) 2016-01-20 2018-03-13 Time Warner Cable Enterprises Llc Apparatus and method for wireless network services in moving vehicles
US9935833B2 (en) 2014-11-05 2018-04-03 Time Warner Cable Enterprises Llc Methods and apparatus for determining an optimized wireless interface installation configuration
US9961383B2 (en) 2008-02-26 2018-05-01 Time Warner Cable Enterprises Llc Methods and apparatus for business-based network resource allocation
US9986578B2 (en) 2015-12-04 2018-05-29 Time Warner Cable Enterprises Llc Apparatus and methods for selective data network access
US9990653B1 (en) * 2014-09-29 2018-06-05 Google Llc Systems and methods for serving online content based on user engagement duration
US20180199088A1 (en) * 2017-01-11 2018-07-12 Invidi Technologies Corporation Managing addressable asset campaigns across multiple devices
US10108983B2 (en) * 2016-01-02 2018-10-23 Facebook, Inc. Selecting content for presentation to an online system user to increase likelihood of user recall of the presented content
US20180316957A1 (en) 2011-10-12 2018-11-01 Turner Broadcasting System, Inc. Advertisement scheduler
FR3067504A1 (en) * 2017-06-16 2018-12-14 Orange METHOD AND DEVICE FOR OBTAINING AN ADVERTISEMENT INTENDED TO BE INSERTED IN DIGITAL CONTENT CONSULTED BY A USER
US10164858B2 (en) 2016-06-15 2018-12-25 Time Warner Cable Enterprises Llc Apparatus and methods for monitoring and diagnosing a wireless network
FR3071085A1 (en) * 2017-09-14 2019-03-15 Amadeus S.A.S. AN AUTOMATIC LEARNING METHOD AND SYSTEM FOR PREDICTING USER INTERACTIONS ONLINE
WO2019052869A1 (en) * 2017-09-14 2019-03-21 Amadeus S.A.S. A machine learning method and system for predicting online user interactions
US10368255B2 (en) 2017-07-25 2019-07-30 Time Warner Cable Enterprises Llc Methods and apparatus for client-based dynamic control of connections to co-existing radio access networks
US10380633B2 (en) 2015-07-02 2019-08-13 The Nielsen Company (Us), Llc Methods and apparatus to generate corrected online audience measurement data
US20190313154A1 (en) * 2018-04-05 2019-10-10 International Business Machines Corporation Customizing digital content based on consumer context data
US10492034B2 (en) 2016-03-07 2019-11-26 Time Warner Cable Enterprises Llc Apparatus and methods for dynamic open-access networks
US10560772B2 (en) 2013-07-23 2020-02-11 Time Warner Cable Enterprises Llc Apparatus and methods for selective data network access
US10567824B2 (en) 2015-09-30 2020-02-18 Simulmedia, Inc. Methods and systems for selecting inventory units for television and other media advertising campaigns
US10586023B2 (en) 2016-04-21 2020-03-10 Time Warner Cable Enterprises Llc Methods and apparatus for secondary content management and fraud prevention
US10636449B2 (en) 2017-11-06 2020-04-28 International Business Machines Corporation Dynamic generation of videos based on emotion and sentiment recognition
US10638361B2 (en) 2017-06-06 2020-04-28 Charter Communications Operating, Llc Methods and apparatus for dynamic control of connections to co-existing radio access networks
US10645547B2 (en) 2017-06-02 2020-05-05 Charter Communications Operating, Llc Apparatus and methods for providing wireless service in a venue
US10755287B2 (en) * 2008-11-25 2020-08-25 Microsoft Technology Licensing, Llc Selecting between client-side and server-side market detection
US10803475B2 (en) 2014-03-13 2020-10-13 The Nielsen Company (Us), Llc Methods and apparatus to compensate for server-generated errors in database proprietor impression data due to misattribution and/or non-coverage
US10834451B2 (en) 2018-01-09 2020-11-10 Turner Broadcasting System, Inc. Dynamically scheduling non-programming media items in contextually relevant programming media content
US20200372502A1 (en) * 2019-05-24 2020-11-26 Blockstack Pbc System and method for smart contract publishing
EP3754996A1 (en) * 2019-06-17 2020-12-23 Accenture Global Solutions Limited Enabling return path data on a non-hybrid set top box for a television
US10909569B1 (en) * 2018-12-09 2021-02-02 Facebook, Inc. Obtaining a composite prediction indicating a likelihood that a user of an online system will perform a conversion associated with a content item via one or more paths of events
US10943184B2 (en) 2017-09-14 2021-03-09 Amadeus S.A.S. Machine learning methods and systems for predicting online user interactions
US11064234B2 (en) 2015-09-01 2021-07-13 Turner Broadcasting System, Inc. Targeting and demographics scheduling utilizing a framework for audience rating estimation
US11082724B2 (en) 2019-08-21 2021-08-03 Dish Network L.L.C. Systems and methods for targeted advertisement insertion into a program content stream
US11093968B2 (en) 2015-11-02 2021-08-17 Turner Broadcasting System, Inc. Audience proposal creation and spot scheduling utilizing a framework for audience rating estimation
US11115710B2 (en) 2017-06-27 2021-09-07 The Nielsen Company (Us), Llc Methods and apparatus to determine synthetic respondent level data using constrained Markov chains
US11120480B2 (en) 2017-09-14 2021-09-14 Amadeus S.A.S. Systems and methods for real-time online traveler segmentation using machine learning
US11140449B2 (en) 2017-02-28 2021-10-05 The Nielsen Company (Us), Llc Methods and apparatus to determine synthetic respondent level data
US11182804B2 (en) * 2016-11-17 2021-11-23 Adobe Inc. Segment valuation in a digital medium environment
US11197050B2 (en) 2013-03-15 2021-12-07 Charter Communications Operating, Llc Methods and apparatus for client-based dynamic control of connections to co-existing radio access networks
US11216834B2 (en) 2019-03-15 2022-01-04 The Nielsen Company (Us), Llc Methods and apparatus to estimate population reach from different marginal ratings and/or unions of marginal ratings based on impression data
US11223860B2 (en) 2007-10-15 2022-01-11 Time Warner Cable Enterprises Llc Methods and apparatus for revenue-optimized delivery of content in a network
US11282115B2 (en) 2017-06-13 2022-03-22 Turner Broadcasting System, Inc. Managing allocation of inventory mix utilizing an optimization framework
US11323772B2 (en) * 2017-02-28 2022-05-03 The Nielsen Company (Us), Llc Methods and apparatus to estimate population reach from different marginal rating unions
US11329902B2 (en) * 2019-03-12 2022-05-10 The Nielsen Company (Us), Llc Methods and apparatus to credit streaming activity using domain level bandwidth information
US11343555B2 (en) 2016-04-05 2022-05-24 Turner Broadcasting System, Inc. Allocation of under delivery units utilizing an optimization framework
US11425458B2 (en) 2017-02-28 2022-08-23 The Nielsen Company (Us), Llc Methods and apparatus to estimate population reach from marginal ratings
US11423431B2 (en) 2017-06-13 2022-08-23 Turner Broadcasting System, Inc. Promotion planning for managing allocation of inventory mix utilizing an optimization framework
US11449268B2 (en) * 2018-11-20 2022-09-20 Samsung Electronics Co., Ltd. Deep solid state device (deep-SSD): a neural network based persistent data storage
US11481802B2 (en) 2020-08-31 2022-10-25 The Nielsen Company (Us), Llc Methods and apparatus for audience and impression deduplication
US11483606B2 (en) 2019-03-15 2022-10-25 The Nielsen Company (Us), Llc Methods and apparatus to estimate population reach from different marginal rating unions
US20220368983A1 (en) * 2013-01-13 2022-11-17 Comcast Cable Communications, Llc Measuring Video-Program-Viewing Activity
US11513815B1 (en) 2019-05-24 2022-11-29 Hiro Systems Pbc Defining data storage within smart contracts
US11523177B2 (en) 2017-02-28 2022-12-06 The Nielsen Company (Us), Llc Methods and apparatus to replicate panelists using a local minimum solution of an integer least squares problem
US11540148B2 (en) 2014-06-11 2022-12-27 Time Warner Cable Enterprises Llc Methods and apparatus for access point location
US11553226B2 (en) 2020-11-16 2023-01-10 The Nielsen Company (Us), Llc Methods and apparatus to estimate population reach from marginal ratings with missing information
US20230032845A1 (en) * 2021-08-02 2023-02-02 The Nielsen Company (Us), Llc Methods and apparatus to generate audience metrics for connected television
US20230079293A1 (en) * 2021-09-15 2023-03-16 Roku, Inc. Demographic Classification of Media Accounts Based on Media Content Data
US11636516B2 (en) 2017-02-13 2023-04-25 Adcuratio Media, Inc. System and method for targeting individuals with advertisement spots during national broadcast and cable television
US11646036B1 (en) * 2022-01-31 2023-05-09 Humancore Llc Team member identification based on psychographic categories
US11657391B1 (en) 2019-05-24 2023-05-23 Hiro Systems Pbc System and method for invoking smart contracts
US20230161745A1 (en) * 2019-07-05 2023-05-25 The Nielsen Company (Us), Llc Methods and apparatus to estimate audience sizes of media using deduplication based on vector of counts sketch data
US11741485B2 (en) 2019-11-06 2023-08-29 The Nielsen Company (Us), Llc Methods and apparatus to estimate de-duplicated unknown total audience sizes based on partial information of known audiences
US11783354B2 (en) 2020-08-21 2023-10-10 The Nielsen Company (Us), Llc Methods and apparatus to estimate census level audience sizes, impression counts, and duration data
US11790397B2 (en) 2021-02-08 2023-10-17 The Nielsen Company (Us), Llc Methods and apparatus to perform computer-based monitoring of audiences of network-based media by using information theory to estimate intermediate level unions
US20230379540A1 (en) * 2019-12-12 2023-11-23 The Nielsen Company (Us), Llc Methods, systems, articles of manufacture and apparatus to remap household identification
US20240061849A1 (en) * 2016-12-02 2024-02-22 Fair Isaac Corporation Fast automatic explanation of scored observations
US11941646B2 (en) 2020-09-11 2024-03-26 The Nielsen Company (Us), Llc Methods and apparatus to estimate population reach from marginals
US11983730B2 (en) 2014-12-31 2024-05-14 The Nielsen Company (Us), Llc Methods and apparatus to correct for deterioration of a demographic model to associate demographic information with media impression information
US12032535B2 (en) 2020-06-30 2024-07-09 The Nielsen Company (Us), Llc Methods and apparatus to estimate audience sizes of media using deduplication based on multiple vectors of counts
US12075103B2 (en) 2010-12-29 2024-08-27 Comcast Cable Communications, Llc Measuring video viewing
US12093968B2 (en) 2020-09-18 2024-09-17 The Nielsen Company (Us), Llc Methods, systems and apparatus to estimate census-level total impression durations and audience size across demographics

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107392220B (en) 2017-05-31 2020-05-05 创新先进技术有限公司 Data stream clustering method and device
US11531080B2 (en) * 2019-07-24 2022-12-20 Cypress Semiconductor Corporation Leveraging spectral diversity for machine learning-based estimation of radio frequency signal parameters

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030130973A1 (en) * 1999-04-05 2003-07-10 American Board Of Family Practice, Inc. Computer architecture and process of patient generation, evolution, and simulation for computer based testing system using bayesian networks as a scripting language
US20040015906A1 (en) * 2001-04-30 2004-01-22 Goraya Tanvir Y. Adaptive dynamic personal modeling system and method
US20070250705A1 (en) * 2006-04-19 2007-10-25 Smith Patrick D Integrated Marketing Portal for Businesses

Family Cites Families (162)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US147156A (en) 1874-02-03 Improvement in sulky-plows
US117250A (en) 1871-07-25 Improvement in the construction and operation of metallurgical furnaces
US33233A (en) 1861-09-10 Improvement in seeding-machines
US33228A (en) 1861-09-10 Improvement in apparatus for injecting meat with brine
US930784A (en) 1909-01-21 1909-08-10 Andrew Cornelius Nelson Agricultural implement.
US1292146A (en) 1917-07-26 1919-01-21 John Taylor Automatic toilet-lock.
US4331974A (en) 1980-10-21 1982-05-25 Iri, Inc. Cable television with controlled signal substitution
US4602279A (en) 1984-03-21 1986-07-22 Actv, Inc. Method for providing targeted profile interactive CATV displays
US4573072A (en) 1984-03-21 1986-02-25 Actv Inc. Method for expanding interactive CATV displayable choices for a given channel capacity
US4918516A (en) 1987-10-26 1990-04-17 501 Actv, Inc. Closed circuit television system having seamless interactive television programming and expandable user participation
US5155591A (en) 1989-10-23 1992-10-13 General Instrument Corporation Method and apparatus for providing demographically targeted television commercials
US5099319A (en) 1989-10-23 1992-03-24 Esch Arthur G Video information delivery method and apparatus
US5446919A (en) 1990-02-20 1995-08-29 Wilkins; Jeff K. Communication system and method with demographically or psychographically defined audiences
US5260778A (en) 1990-06-26 1993-11-09 General Instrument Corporation Apparatus for selective distribution of messages over a communications network
US5231494A (en) 1991-10-08 1993-07-27 General Instrument Corporation Selection of compressed television signals from single channel allocation based on viewer characteristics
US5724091A (en) 1991-11-25 1998-03-03 Actv, Inc. Compressed digital data interactive program system
FI92139C (en) 1992-02-28 1994-10-10 Matti Myllymaeki Monitoring device for the health condition, which is attached to the wrist
US5537586A (en) 1992-04-30 1996-07-16 Individual, Inc. Enhanced apparatus and methods for retrieving and selecting profiled textural information records from a database of defined category structures
DE4226990C3 (en) 1992-08-14 1999-04-29 Sirona Dental Systems Gmbh Video camera for viewing objects in a patient's mouth
WO1994014284A1 (en) 1992-12-09 1994-06-23 Discovery Communications, Inc. Reprogrammable terminal for suggesting programs offered on a television program delivery system
US5600364A (en) 1992-12-09 1997-02-04 Discovery Communications, Inc. Network controller for cable television delivery systems
US6463585B1 (en) 1992-12-09 2002-10-08 Discovery Communications, Inc. Targeted advertisement using television delivery systems
US5381477A (en) 1993-02-16 1995-01-10 Scientific-Atlanta, Inc. Method of selecting cable television converter groups
US5410344A (en) 1993-09-22 1995-04-25 Arrowsmith Technologies, Inc. Apparatus and method of selecting video programs based on viewers' preferences
EP0656729B1 (en) 1993-12-01 2000-02-09 Matsushita Electric Industrial Co., Ltd. Method and apparatus for editing or mixing compressed pictures
US5537141A (en) 1994-04-15 1996-07-16 Actv, Inc. Distance learning system providing individual television participation, audio responses and memory for every student
US5636346A (en) 1994-05-09 1997-06-03 The Electronic Address, Inc. Method and system for selectively targeting advertisements and programming
US5534941A (en) 1994-05-20 1996-07-09 Encore Media Corporation System for dynamic real-time television channel expansion
US5534944A (en) 1994-07-15 1996-07-09 Matsushita Electric Corporation Of America Method of splicing MPEG encoded video
US5515098A (en) 1994-09-08 1996-05-07 Carles; John B. System and method for selectively distributing commercial messages over a communications network
US5724521A (en) 1994-11-03 1998-03-03 Intel Corporation Method and apparatus for providing electronic advertisements to end users in a consumer best-fit pricing manner
US5758257A (en) 1994-11-29 1998-05-26 Herz; Frederick System and method for scheduling broadcast of and access to video programs and other data using customer profiles
US5774170A (en) 1994-12-13 1998-06-30 Hite; Kenneth C. System and method for delivering targeted advertisements to consumers
US5600366A (en) 1995-03-22 1997-02-04 Npb Partners, Ltd. Methods and apparatus for digital advertisement insertion in video programming
US5740549A (en) 1995-06-12 1998-04-14 Pointcast, Inc. Information and advertising distribution system and method
US5652615A (en) 1995-06-30 1997-07-29 Digital Equipment Corporation Precision broadcast of composite programs including secondary program content such as advertisements
US5805974A (en) 1995-08-08 1998-09-08 Hite; Kenneth C. Method and apparatus for synchronizing commercial advertisements across multiple communication channels
US6002393A (en) 1995-08-22 1999-12-14 Hite; Kenneth C. System and method for delivering targeted advertisements to consumers using direct commands
WO1997016184A1 (en) 1995-11-02 1997-05-09 Warner-Lambert Company Method and pharmaceutical composition for regulating lipid concentration
US5872588A (en) 1995-12-06 1999-02-16 International Business Machines Corporation Method and apparatus for monitoring audio-visual materials presented to a subscriber
WO1997026729A2 (en) 1995-12-27 1997-07-24 Robinson Gary B Automated collaborative filtering in world wide web advertising
US5859660A (en) 1996-02-29 1999-01-12 Perkins; Michael G. Non-seamless splicing of audio-video transport streams
US5818539A (en) 1996-03-29 1998-10-06 Matsushita Electric Corporation Of America System and method for updating a system time constant (STC) counter following a discontinuity in an MPEG-2 transport data stream
US5848396A (en) 1996-04-26 1998-12-08 Freedom Of Information, Inc. Method and apparatus for determining behavioral profile of a computer user
US5838678A (en) 1996-07-24 1998-11-17 Davis; Joseph W. Method and device for preprocessing streams of encoded data to facilitate decoding streams back-to back
US5917830A (en) 1996-10-18 1999-06-29 General Instrument Corporation Splicing compressed packetized digital video streams
US5948061A (en) 1996-10-29 1999-09-07 Double Click, Inc. Method of delivery, targeting, and measuring advertising over networks
DE69715040T2 (en) 1996-12-20 2003-05-08 Princeton Video Image, Inc. ADAPTER FOR TARGETED ELECTRONIC INSERTION OF CHARACTERS IN VIDEO SIGNALS
EP0965192B1 (en) 1997-01-06 2007-02-28 Bellsouth Intellectual Property Corporation Method and system for tracking network use
US6067303A (en) 1997-02-25 2000-05-23 International Business Machines Corporation Method and apparatus for detecting and controlling data stream splicing in ATM networks
US5982436A (en) 1997-03-28 1999-11-09 Philips Electronics North America Corp. Method for seamless splicing in a video encoder
US6151443A (en) 1997-05-16 2000-11-21 Indigita Corporation Digital video and data recorder
US6353929B1 (en) 1997-06-23 2002-03-05 One River Worldtrek, Inc. Cooperative system for measuring electronic media
GB9721947D0 (en) 1997-10-16 1997-12-17 Thomson Consumer Electronics Intelligent IP packet scheduler algorithm
WO1999021338A1 (en) 1997-10-16 1999-04-29 Thomson Licensing S.A. System for prioritizing bi-directional broadcast data
US6154496A (en) 1997-11-25 2000-11-28 Philips Electronics N.A. Corp. Video buffer for seamless splicing of MPEG streams
US6029045A (en) 1997-12-09 2000-02-22 Cogent Technology, Inc. System and method for inserting local content into programming content
US6049569A (en) 1997-12-09 2000-04-11 Philips Electronics N.A. Corporation Method and apparatus for encoding digital video bit streams with seamless splice points and method and apparatus for splicing such digital video bit streams
US6195368B1 (en) 1998-01-14 2001-02-27 Skystream Corporation Re-timing of video program bearing streams transmitted by an asynchronous communication link
US6246701B1 (en) 1998-01-14 2001-06-12 Skystream Corporation Reference time clock locking in a remultiplexer for video program bearing transport streams
US6111896A (en) 1998-01-14 2000-08-29 Skystream Corporation Remultiplexer for video program bearing transport streams with program clock reference time stamp adjustment
US6487721B1 (en) 1998-01-30 2002-11-26 General Instrument Corporation Apparatus and method for digital advertisement insertion in a bitstream
WO1999046708A1 (en) 1998-03-11 1999-09-16 Iq.Commerce Corporation Method and system for delivering and redeeming dynamically and adaptively characterized promotional incentives on a computer network
US6269120B1 (en) 1998-03-23 2001-07-31 International Business Machines Corporation Method of precise buffer management for MPEG video splicing
US6252873B1 (en) 1998-06-17 2001-06-26 Gregory O. Vines Method of ensuring a smooth transition between MPEG-2 transport streams
CA2345241A1 (en) 1998-09-22 2000-03-30 Science Applications International Corporation User-defined dynamic collaborative environments
US6317881B1 (en) 1998-11-04 2001-11-13 Intel Corporation Method and apparatus for collecting and providing viewer feedback to a broadcast
KR20010080633A (en) 1998-11-30 2001-08-22 추후제출 Smart agent based on habit, statistical inference and psycho-demographic profiling
US8180675B2 (en) 2000-08-31 2012-05-15 Prime Research Alliance E., Inc. System and method for automatically managing avail inventory data and avail pricing
US7228555B2 (en) 2000-08-31 2007-06-05 Prime Research Alliance E., Inc. System and method for delivering targeted advertisements using multiple presentation streams
US20020072966A1 (en) 2000-08-31 2002-06-13 Eldering Charles A. System for providing targeted advertisements using advertiser-specific target groups
US6560578B2 (en) 1999-03-12 2003-05-06 Expanse Networks, Inc. Advertisement selection system supporting discretionary target market characteristics
US6820277B1 (en) 1999-04-20 2004-11-16 Expanse Networks, Inc. Advertising management system for digital video streams
US6684194B1 (en) 1998-12-03 2004-01-27 Expanse Network, Inc. Subscriber identification system
US7185353B2 (en) 2000-08-31 2007-02-27 Prime Research Alliance E., Inc. System and method for delivering statistically scheduled advertisements
US6298348B1 (en) 1998-12-03 2001-10-02 Expanse Networks, Inc. Consumer profiling system
US20020083439A1 (en) 2000-08-31 2002-06-27 Eldering Charles A. System for rescheduling and inserting advertisements
US20010049620A1 (en) 2000-02-29 2001-12-06 Blasko John P. Privacy-protected targeting system
US20020083445A1 (en) 2000-08-31 2002-06-27 Flickinger Gregory C. Delivering targeted advertisements to the set-top-box
US6216129B1 (en) 1998-12-03 2001-04-10 Expanse Networks, Inc. Advertisement selection system supporting discretionary target market characteristics
US6704930B1 (en) 1999-04-20 2004-03-09 Expanse Networks, Inc. Advertisement insertion techniques for digital video streams
US6457010B1 (en) 1998-12-03 2002-09-24 Expanse Networks, Inc. Client-server based subscriber characterization system
US20020111172A1 (en) 2001-02-14 2002-08-15 Dewolf Frederik M. Location based profiling
US6324519B1 (en) 1999-03-12 2001-11-27 Expanse Networks, Inc. Advertisement auction system
US20020087973A1 (en) 2000-12-28 2002-07-04 Hamilton Jeffrey S. Inserting local signals during MPEG channel changes
US7039932B2 (en) 2000-08-31 2006-05-02 Prime Research Alliance E., Inc. Queue-based head-end advertisement scheduling method and apparatus
US7260823B2 (en) 2001-01-11 2007-08-21 Prime Research Alliance E., Inc. Profiling and identification of television viewers
US8290351B2 (en) 2001-04-03 2012-10-16 Prime Research Alliance E., Inc. Alternative advertising in prerecorded media
US7328448B2 (en) 2000-08-31 2008-02-05 Prime Research Alliance E, Inc. Advertisement distribution system for distributing targeted advertisements in television systems
US7653923B2 (en) 2000-02-18 2010-01-26 Prime Research Alliance E, Inc. Scheduling and presenting IPG ads in conjunction with programming ads in a television environment
US20020083441A1 (en) 2000-08-31 2002-06-27 Flickinger Gregory C. Advertisement filtering and storage for targeted advertisement systems
US20020026638A1 (en) 2000-08-31 2002-02-28 Eldering Charles A. Internet-based electronic program guide advertisement insertion method and apparatus
US20020144263A1 (en) 2000-08-31 2002-10-03 Eldering Charles A. Grouping of advertisements on an advertising channel in a targeted advertisement system
US20020123928A1 (en) 2001-01-11 2002-09-05 Eldering Charles A. Targeting ads to subscribers based on privacy-protected subscriber profiles
ES2397354T3 (en) 1999-02-17 2013-03-06 Index Systems, Inc. System and method to adapt features of television and / or electronic programming guide, such as advertising
US6813775B1 (en) 1999-03-29 2004-11-02 The Directv Group, Inc. Method and apparatus for sharing viewing preferences
US8943527B2 (en) 1999-03-30 2015-01-27 Tivo Inc. Audience measurement system
EP1177674A4 (en) 1999-05-10 2004-12-08 Samsung Electronics Co Ltd Advertisement subgroups for digital streams
US6438752B1 (en) 1999-06-22 2002-08-20 Mediaone Group, Inc. Method and system for selecting television programs based on the past selection history of an identified user
US6708335B1 (en) 1999-08-18 2004-03-16 Webtv Networks, Inc. Tracking viewing behavior of advertisements on a home entertainment system
US20020099611A1 (en) 1999-12-06 2002-07-25 De Souza Celso Candido Formation of horizontal, vertical and diagonal databases in an extranet based e-commerce platform
CA2394483A1 (en) 1999-12-17 2001-06-21 Technology, Patents And Licensing, Inc. Global asset information registry
US7100183B2 (en) 2000-02-02 2006-08-29 Sedna Patent Services, Llc System and method for transmitting and displaying targeted infromation
US8572646B2 (en) 2000-04-07 2013-10-29 Visible World Inc. System and method for simultaneous broadcast for personalized messages
US20040148625A1 (en) 2000-04-20 2004-07-29 Eldering Charles A Advertisement management system for digital video streams
AU2001264676A1 (en) 2000-05-19 2001-12-03 Intellibridge Corporation Method and apparatus for providing customized information
CA2349914C (en) 2000-06-09 2013-07-30 Invidi Technologies Corp. Advertising delivery method
US8495679B2 (en) 2000-06-30 2013-07-23 Thomson Licensing Method and apparatus for delivery of television programs and targeted de-coupled advertising
JP4240182B2 (en) 2000-08-09 2009-03-18 株式会社エヌ・ティ・ティ・データ Advertisement exposure management method and receiver
US20020083435A1 (en) 2000-08-31 2002-06-27 Blasko John P. Method and system for addressing targeted advertisements using detection of operational status of display device
WO2002019717A2 (en) 2000-08-31 2002-03-07 Myrio Corporation Real-time audience monitoring, content rating, and content enhancing
US7209921B2 (en) 2000-09-01 2007-04-24 Op40, Inc. Method and system for deploying an asset over a multi-tiered network
US20020092019A1 (en) 2000-09-08 2002-07-11 Dwight Marcus Method and apparatus for creation, distribution, assembly and verification of media
US8020183B2 (en) 2000-09-14 2011-09-13 Sharp Laboratories Of America, Inc. Audiovisual management system
US8302127B2 (en) 2000-09-25 2012-10-30 Thomson Licensing System and method for personalized TV
US7490344B2 (en) 2000-09-29 2009-02-10 Visible World, Inc. System and method for seamless switching
US20020083451A1 (en) 2000-12-21 2002-06-27 Gill Komlika K. User-friendly electronic program guide based on subscriber characterizations
IL156146A0 (en) 2000-11-28 2003-12-23 Seachange International Inc Content/service handling and delivery
US7331057B2 (en) 2000-12-28 2008-02-12 Prime Research Alliance E, Inc. Grouping advertisement subavails
ATE321422T1 (en) 2001-01-09 2006-04-15 Metabyte Networks Inc SYSTEM, METHOD AND SOFTWARE FOR PROVIDING TARGETED ADVERTISING THROUGH USER PROFILE DATA STRUCTURE BASED ON USER PREFERENCES
US20020111154A1 (en) 2001-02-14 2002-08-15 Eldering Charles A. Location based delivery
US20020178445A1 (en) 2001-04-03 2002-11-28 Charles Eldering Subscriber selected advertisement display and scheduling
US20020178447A1 (en) 2001-04-03 2002-11-28 Plotnick Michael A. Behavioral targeted advertising
US20020184047A1 (en) 2001-04-03 2002-12-05 Plotnick Michael A. Universal ad queue
US20020143901A1 (en) 2001-04-03 2002-10-03 Gtech Rhode Island Corporation Interactive media response processing system
US20020184130A1 (en) 2001-04-04 2002-12-05 Blasko John P. Intellectual capital risk analysis system
US20030005437A1 (en) 2001-06-01 2003-01-02 Norman Feuer Networked broadcasting system with demographically controlled advertisement selection
US7730509B2 (en) 2001-06-08 2010-06-01 Invidi Technologies Corporation Asset delivery reporting in a broadcast network
US20040015984A1 (en) 2001-06-29 2004-01-22 Tomoyuki Yamamoto Information providing system, information provding de and method, information processing device and method
US7657428B2 (en) 2001-07-09 2010-02-02 Visible World, Inc System and method for seamless switching of compressed audio streams
US20030018584A1 (en) 2001-07-23 2003-01-23 Cohen Jeremy Stein System and method for analyzing transaction data
US20030097299A1 (en) 2001-11-21 2003-05-22 O'kane Robert Peer-to-peer (P2P) and internet content digital acknowledgement trigger used for file transfer
US7136871B2 (en) 2001-11-21 2006-11-14 Microsoft Corporation Methods and systems for selectively displaying advertisements
US20030101454A1 (en) 2001-11-21 2003-05-29 Stuart Ozer Methods and systems for planning advertising campaigns
US20030105831A1 (en) 2001-12-04 2003-06-05 O'kane Robert Peer-to-peer (P2P) and internet content delivery based user based digital acknowledgement trigger used for file transfer
US20050021398A1 (en) 2001-11-21 2005-01-27 Webhound Corporation Method and system for downloading digital content over a network
US20030115318A1 (en) 2001-12-13 2003-06-19 Microsoft Corporation. Concentric user-targeting delivery system and methods
US6978470B2 (en) 2001-12-26 2005-12-20 Bellsouth Intellectual Property Corporation System and method for inserting advertising content in broadcast programming
US20030149975A1 (en) 2002-02-05 2003-08-07 Charles Eldering Targeted advertising in on demand programming
WO2004019530A1 (en) 2002-02-15 2004-03-04 Visible World, Inc. System and method for seamless switching through buffering
US20030200336A1 (en) 2002-02-15 2003-10-23 Suparna Pal Apparatus and method for the delivery of multiple sources of media content
US20030204592A1 (en) 2002-03-07 2003-10-30 Crown Media Holdings, Inc. System for uniquely identifying assets and subsribers in a multi-media communicaion network
US9235849B2 (en) 2003-12-31 2016-01-12 Google Inc. Generating user information for use in targeted advertising
US7174126B2 (en) 2002-05-03 2007-02-06 Time Warner Interactive Video Group Inc. Technique for effectively accessing programming listing information in an entertainment delivery system
AU2003267975A1 (en) 2002-06-27 2004-01-19 Piranha Media Distribution, Inc. Method and apparatus for the free licensing of digital media content
US8028092B2 (en) 2002-06-28 2011-09-27 Aol Inc. Inserting advertising content
US20040045020A1 (en) 2002-08-29 2004-03-04 Witt Jeffrey Michael Commercial identification system
AU2003270459A1 (en) 2002-09-06 2004-03-29 Visible World, Inc. System for authoring and editing personalized message campaigns
US7398261B2 (en) 2002-11-20 2008-07-08 Radar Networks, Inc. Method and system for managing and tracking semantic objects
JP2004206324A (en) 2002-12-25 2004-07-22 Individual Network Llc Method and system for ordering advertisement spot through data network
US20040173311A1 (en) 2003-03-04 2004-09-09 Tomoyoshi Ichimaru Plasma processing apparatus and method
US7448058B2 (en) 2003-03-10 2008-11-04 Mark A. Heyner Method for enabling a television user to control operation of application programs on a programmable television controller
US7188048B2 (en) 2003-06-25 2007-03-06 Lockheed Martin Corporation Refining stochastic grid filter
US20050080846A1 (en) 2003-09-27 2005-04-14 Webhound, Inc. Method and system for updating digital content over a network
US9396212B2 (en) 2004-04-07 2016-07-19 Visible World, Inc. System and method for enhanced video selection
US20050234992A1 (en) 2004-04-07 2005-10-20 Seth Haberman Method and system for display guide for video selection
KR20080043764A (en) * 2005-06-28 2008-05-19 초이스스트림, 인코포레이티드 Methods and apparatus for a statistical system for targeting advertisements
US8713607B2 (en) 2005-09-30 2014-04-29 Microsoft Corporation Multi-room user interface
US20070162331A1 (en) * 2006-01-10 2007-07-12 Michael Sullivan Systems and methods for providing information and conducting business using the internet
US20070239675A1 (en) 2006-03-29 2007-10-11 Microsoft Corporation Web search media service
CN107103489B (en) 2006-05-02 2021-09-28 英维迪技术公司 Method and device for determining target for useful resource in broadcast network
JP4506773B2 (en) 2007-03-28 2010-07-21 ブラザー工業株式会社 Electrical connection between both sides of the board

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030130973A1 (en) * 1999-04-05 2003-07-10 American Board Of Family Practice, Inc. Computer architecture and process of patient generation, evolution, and simulation for computer based testing system using bayesian networks as a scripting language
US20040015906A1 (en) * 2001-04-30 2004-01-22 Goraya Tanvir Y. Adaptive dynamic personal modeling system and method
US20070250705A1 (en) * 2006-04-19 2007-10-25 Smith Patrick D Integrated Marketing Portal for Businesses

Cited By (151)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11223860B2 (en) 2007-10-15 2022-01-11 Time Warner Cable Enterprises Llc Methods and apparatus for revenue-optimized delivery of content in a network
US9961383B2 (en) 2008-02-26 2018-05-01 Time Warner Cable Enterprises Llc Methods and apparatus for business-based network resource allocation
US20200402074A1 (en) * 2008-11-25 2020-12-24 Microsoft Technology Licensing, Llc Selecting between client-side and server-side market detection
US11669850B2 (en) * 2008-11-25 2023-06-06 Microsoft Technology Licensing, Llc Selecting between client-side and server-side market detection
US10755287B2 (en) * 2008-11-25 2020-08-25 Microsoft Technology Licensing, Llc Selecting between client-side and server-side market detection
US12075103B2 (en) 2010-12-29 2024-08-27 Comcast Cable Communications, Llc Measuring video viewing
US20180316957A1 (en) 2011-10-12 2018-11-01 Turner Broadcasting System, Inc. Advertisement scheduler
US10701423B2 (en) 2011-10-12 2020-06-30 Turner Broadcasting Systems, Inc. Advertisement scheduler
US9788149B2 (en) 2012-08-30 2017-10-10 Time Warner Cable Enterprises Llc Apparatus and methods for enabling location-based services within a premises
US10278008B2 (en) 2012-08-30 2019-04-30 Time Warner Cable Enterprises Llc Apparatus and methods for enabling location-based services within a premises
US10715961B2 (en) 2012-08-30 2020-07-14 Time Warner Cable Enterprises Llc Apparatus and methods for enabling location-based services within a premises
US11968421B2 (en) * 2013-01-13 2024-04-23 Comcast Cable Communications, Llc Measuring video-program-viewing activity
US20220368983A1 (en) * 2013-01-13 2022-11-17 Comcast Cable Communications, Llc Measuring Video-Program-Viewing Activity
US11197050B2 (en) 2013-03-15 2021-12-07 Charter Communications Operating, Llc Methods and apparatus for client-based dynamic control of connections to co-existing radio access networks
US9894498B2 (en) * 2013-03-16 2018-02-13 Wellpepper, Inc. Adaptive notification system for mobile devices
US20170150331A1 (en) * 2013-03-16 2017-05-25 Wellpepper, Inc. Adaptive notification system for mobile devices
US10560772B2 (en) 2013-07-23 2020-02-11 Time Warner Cable Enterprises Llc Apparatus and methods for selective data network access
US11425076B1 (en) * 2013-10-30 2022-08-23 Mesh Labs Inc. Method and system for filtering electronic communications
US10805251B2 (en) * 2013-10-30 2020-10-13 Mesh Labs Inc. Method and system for filtering electronic communications
US20150120848A1 (en) * 2013-10-30 2015-04-30 Mesh Labs Inc. Method and system for filtering electronic communications
US9277278B2 (en) * 2014-02-20 2016-03-01 Google Inc. Systems and methods for enhancing audience measurement data
US8910195B1 (en) * 2014-02-20 2014-12-09 Google Inc. Systems and methods for enhancing audience measurement data
US20150237395A1 (en) * 2014-02-20 2015-08-20 Google Inc. Systems and methods for enhancing audience measurement data
US10803475B2 (en) 2014-03-13 2020-10-13 The Nielsen Company (Us), Llc Methods and apparatus to compensate for server-generated errors in database proprietor impression data due to misattribution and/or non-coverage
US11568431B2 (en) 2014-03-13 2023-01-31 The Nielsen Company (Us), Llc Methods and apparatus to compensate for server-generated errors in database proprietor impression data due to misattribution and/or non-coverage
US12045845B2 (en) 2014-03-13 2024-07-23 The Nielsen Company (Us), Llc Methods and apparatus to compensate for server-generated errors in database proprietor impression data due to misattribution and/or non-coverage
WO2015139119A1 (en) * 2014-03-19 2015-09-24 Verosource Solutions Inc. System and method for validating data source input from a crowd sourcing platform
EP2945113A1 (en) * 2014-05-14 2015-11-18 Cisco Technology, Inc. Audience segmentation using machine-learning
US11540148B2 (en) 2014-06-11 2022-12-27 Time Warner Cable Enterprises Llc Methods and apparatus for access point location
US10028025B2 (en) * 2014-09-29 2018-07-17 Time Warner Cable Enterprises Llc Apparatus and methods for enabling presence-based and use-based services
US11544741B2 (en) * 2014-09-29 2023-01-03 Google Llc Systems and methods for serving online content based on user engagement duration
US20160094883A1 (en) * 2014-09-29 2016-03-31 Time Warner Cable Enterprises Llc Apparatus and methods for enabling presence-based and use-based services
US11082743B2 (en) 2014-09-29 2021-08-03 Time Warner Cable Enterprises Llc Apparatus and methods for enabling presence-based and use-based services
US10949878B2 (en) 2014-09-29 2021-03-16 Google Llc Systems and methods for serving online content based on user engagement duration
US9990653B1 (en) * 2014-09-29 2018-06-05 Google Llc Systems and methods for serving online content based on user engagement duration
US20160094894A1 (en) * 2014-09-30 2016-03-31 Nbcuniversal Media, Llc Digital content audience matching and targeting system and method
US10834450B2 (en) * 2014-09-30 2020-11-10 Nbcuniversal Media, Llc Digital content audience matching and targeting system and method
US9935833B2 (en) 2014-11-05 2018-04-03 Time Warner Cable Enterprises Llc Methods and apparatus for determining an optimized wireless interface installation configuration
US11983730B2 (en) 2014-12-31 2024-05-14 The Nielsen Company (Us), Llc Methods and apparatus to correct for deterioration of a demographic model to associate demographic information with media impression information
US20190370860A1 (en) * 2015-07-02 2019-12-05 The Nielsen Company (Us), Llc Methods and apparatus to generate corrected online audience measurement data
US11706490B2 (en) 2015-07-02 2023-07-18 The Nielsen Company (Us), Llc Methods and apparatus to correct errors in audience measurements for media accessed using over-the-top devices
US12015826B2 (en) 2015-07-02 2024-06-18 The Nielsen Company (Us), Llc Methods and apparatus to correct errors in audience measurements for media accessed using over-the-top devices
US10045082B2 (en) * 2015-07-02 2018-08-07 The Nielsen Company (Us), Llc Methods and apparatus to correct errors in audience measurements for media accessed using over-the-top devices
US11259086B2 (en) 2015-07-02 2022-02-22 The Nielsen Company (Us), Llc Methods and apparatus to correct errors in audience measurements for media accessed using over the top devices
US10368130B2 (en) * 2015-07-02 2019-07-30 The Nielsen Company (Us), Llc Methods and apparatus to correct errors in audience measurements for media accessed using over the top devices
US20170006342A1 (en) * 2015-07-02 2017-01-05 The Nielsen Company (Us), Llc Methods and apparatus to correct errors in audience measurements for media accessed using over-the-top devices
US10785537B2 (en) 2015-07-02 2020-09-22 The Nielsen Company (Us), Llc Methods and apparatus to correct errors in audience measurements for media accessed using over the top devices
US11645673B2 (en) * 2015-07-02 2023-05-09 The Nielsen Company (Us), Llc Methods and apparatus to generate corrected online audience measurement data
US10380633B2 (en) 2015-07-02 2019-08-13 The Nielsen Company (Us), Llc Methods and apparatus to generate corrected online audience measurement data
US11064234B2 (en) 2015-09-01 2021-07-13 Turner Broadcasting System, Inc. Targeting and demographics scheduling utilizing a framework for audience rating estimation
US9838754B2 (en) 2015-09-01 2017-12-05 The Nielsen Company (Us), Llc On-site measurement of over the top media
US10567824B2 (en) 2015-09-30 2020-02-18 Simulmedia, Inc. Methods and systems for selecting inventory units for television and other media advertising campaigns
US20170127110A1 (en) * 2015-11-02 2017-05-04 Turner Broadcasting System, Inc. Generation of reach, mixture, and pricing utilizing a framework for audence rating estimation
US11093968B2 (en) 2015-11-02 2021-08-17 Turner Broadcasting System, Inc. Audience proposal creation and spot scheduling utilizing a framework for audience rating estimation
US10070166B2 (en) * 2015-11-02 2018-09-04 Turner Broadcasting System, Inc Generation of reach, mixture, and pricing utilizing a framework for audience rating estimation
US11412320B2 (en) 2015-12-04 2022-08-09 Time Warner Cable Enterprises Llc Apparatus and methods for selective data network access
US9986578B2 (en) 2015-12-04 2018-05-29 Time Warner Cable Enterprises Llc Apparatus and methods for selective data network access
US10755311B1 (en) * 2016-01-02 2020-08-25 Facebook, Inc. Selecting content for presentation to an online system user to increase likelihood of user recall of the presented content
US10108983B2 (en) * 2016-01-02 2018-10-23 Facebook, Inc. Selecting content for presentation to an online system user to increase likelihood of user recall of the presented content
US11778285B2 (en) 2016-01-05 2023-10-03 Roku, Inc. Computing system with channel-change-based trigger feature
US12081845B2 (en) 2016-01-05 2024-09-03 Roku, Inc. Computing system with channel-change-based trigger feature
US20170195714A1 (en) * 2016-01-05 2017-07-06 Gracenote, Inc. Computing System with Channel-Change-Based Trigger Feature
US10939185B2 (en) * 2016-01-05 2021-03-02 Gracenote, Inc. Computing system with channel-change-based trigger feature
US20170208353A1 (en) * 2016-01-20 2017-07-20 Samsung Electronics Co., Ltd. Content selection for networked media devices
US9918345B2 (en) 2016-01-20 2018-03-13 Time Warner Cable Enterprises Llc Apparatus and method for wireless network services in moving vehicles
US11019385B2 (en) * 2016-01-20 2021-05-25 Samsung Electronics Co., Ltd. Content selection for networked media devices
US10687371B2 (en) 2016-01-20 2020-06-16 Time Warner Cable Enterprises Llc Apparatus and method for wireless network services in moving vehicles
US20210357435A1 (en) * 2016-02-03 2021-11-18 Mx Technologies, Inc. Automated data supplementation and verification
US11093528B2 (en) * 2016-02-03 2021-08-17 Mx Technologies, Inc. Automated data supplementation and verification
US20170220670A1 (en) * 2016-02-03 2017-08-03 Mx Technologies, Inc. Automated data supplementation and verification
US11665509B2 (en) 2016-03-07 2023-05-30 Time Warner Cable Enterprises Llc Apparatus and methods for dynamic open-access networks
US10492034B2 (en) 2016-03-07 2019-11-26 Time Warner Cable Enterprises Llc Apparatus and methods for dynamic open-access networks
US9872072B2 (en) 2016-03-21 2018-01-16 Google Llc Systems and methods for identifying non-canonical sessions
US10575054B2 (en) 2016-03-21 2020-02-25 Google Llc. Systems and methods for identifying non-canonical sessions
US11343555B2 (en) 2016-04-05 2022-05-24 Turner Broadcasting System, Inc. Allocation of under delivery units utilizing an optimization framework
US12081819B2 (en) 2016-04-05 2024-09-03 Turner Broadcasting System, Inc. Allocation of under delivery units utilizing an optimization framework
US10586023B2 (en) 2016-04-21 2020-03-10 Time Warner Cable Enterprises Llc Methods and apparatus for secondary content management and fraud prevention
US11669595B2 (en) 2016-04-21 2023-06-06 Time Warner Cable Enterprises Llc Methods and apparatus for secondary content management and fraud prevention
US11146470B2 (en) 2016-06-15 2021-10-12 Time Warner Cable Enterprises Llc Apparatus and methods for monitoring and diagnosing a wireless network
US10164858B2 (en) 2016-06-15 2018-12-25 Time Warner Cable Enterprises Llc Apparatus and methods for monitoring and diagnosing a wireless network
US11182804B2 (en) * 2016-11-17 2021-11-23 Adobe Inc. Segment valuation in a digital medium environment