CN113888207A - Revenue optimization for cross-screen advertising - Google Patents

Abstract

The present invention relates to a computer-implemented method for optimizing placement of advertising content on a plurality of different devices. The system allows for targeting advertising content to consumers on televisions and mobile devices, which may operate at a multi-channel video program distributor. The system can exploit hard and soft constraints to set some suitable goals for an advertising campaign and can provide tools to optimize these goals. The system may assign advertising campaigns and plans to different inventory types based on the probability of an exact match to a consumer. Consumer matching may be achieved by generating similar models in the consumer device graph to predict future consumption behavior. The system includes an interface through which the user can adjust various constraints and optimize the revenue the distributor receives from the advertisement.

Description

Revenue optimization for cross-screen advertising

Divisional application

The application is a divisional application with the application number of 201780016959.6, application date of 2017, 1 month and 13 days, and the title of "revenue optimization of cross-screen advertisement delivery".

Priority

This application claims priority from 62/278,888 filed 2016/14/2016 and U.S. provisional application No. 62/290,387 filed 2016/2/2016, and is a continuation of U.S. patent application No. 15/219,262 filed 2016/7/25/2016, according to 35U.S. C. 119(e), the entire contents of which are incorporated herein by reference in their entirety.

RELATED APPLICATIONS

This application is related to the following U.S. patent applications: U.S. patent application Ser. No. 15/219,259 entitled "TARGETING TV ADVERTISING SLOTS BASED ON CONSUMER ONLINE BEHAVIOR", filed 7/25/2016; U.S. patent application Ser. No. 15/219,268 entitled CROSS-SCREEN MEASUREMENT ACCURACY IN ADVERTISING PERFOMANCE filed on 25/7/2016; U.S. patent application Ser. No. 15/219,264 entitled "SEQUENTIAL DELIVERY OF ADVERTISING CONTENT ACROSS MEDIA DEVICES", filed 2016, 7, 25.D.; and serial No. 62/317,440 provisional application filed on 2016, 4, month, 2; the entire contents of which are hereby incorporated by reference in their entirety.

Technical Field

The technology described herein relates generally to improvements and management of cross-screen advertising schemes for advertisers, and more particularly to a system and method for targeting advertising content to consumers on television and mobile devices, which may operate in the environment of a multi-channel video program distributor.

Background

Video advertisements are one of the most advanced, complex and expensive forms of advertising content, particularly those intended for delivery to more and more virtual reality devices, and those that require some degree of user interactivity to achieve their desired effect. In addition to the cost of producing the video content itself, the cost of delivering video content over broadcast and cable television networks remains significant, in part because Television (TV) slots are a premium advertising space in today's economy. Furthermore, televisions are no longer the entirety of the media market. Consumers can now propagate their mindsets for video content, especially premium content viewed through televisions, DVRs, and animal shows beyond peak and on-demand video services viewed through smart televisions, game consoles, mobile devices, and traditional televisions.

In short, television viewing is transitioning to digital distributed viewing because the live broadcast of viewer viewing is relatively small, while the amount of viewing in video-on-demand (VOD) or streaming video formats is relatively large. Accordingly, broadcasters are increasingly able to more accurately determine specific market segments, thereby providing opportunities for advertisers who purchase time on their channels to target their content to highly relevant groups.

Adding online consumption to the list of options available to any given consumer only results in a more complex integration process for delivering video advertisements to the relevant public segment, at least in part because of the number of different types of devices that may be online now and in the future, and the many different types of online devices that an individual may interact with on any one day. This complexity means that the task of optimizing ad content placement today far exceeds what has traditionally been necessary, and beyond what has previously been done by experienced people. It is sporadic to have a full understanding of the data needed for a given consumer, as each individual and household views more and more media in different ways by accessing network devices.

In addition, the advertisement scheduling problem increases as the number of advertisements and inventory slots increase. As such, existing approaches to preventing accidental violations of constraints on ad placement (i.e., media content administration or contractual obligations) are very ineffective, even for a limited number of constraints. While existing advertising product optimization software tools allow constraints to be explored through a simple relational database, they are unable to analyze and suggest advertising layouts based on these constraints.

For many companies, today's analysis work to study advertising strategies still requires the manual assistance of analysts. This is particularly true for purchasers with low numbers of advertising inventory. The advertising strategy is also typically fixed, meaning that the approach of the advertising strategy depends on certain assumptions, which are inflexible and limited to manual approaches.

Assigning values to soft constraints, such as business goals and internal vendor preferences of existing advertiser customers, is also a challenge. The system and method cannot explore various weighting methods for selling advertising inventory because it is too time consuming and expensive to do so because it requires manual modeling on Excel style spreadsheets. Furthermore, in the current market, new sources of demand can be added in real time, and the complexity of such new potential activities cannot be addressed by manual or spreadsheet methods. The market is real time and running at all times.

The current state of the advertising strategy is similar to the situation before financial strategic instruments (such as E-trading that facilitates automated purchases) and financial consultants (such as the Fidelity investment program) are present.

In today's advertising strategies, manual analysis guides the selection of advertising inventory based on, for example, EXCEL data sheets and other static data management tools. This results in inefficiency in selecting a gear and delays in reacting to market trends. Consumer preferences are not completely different depending on the device used by the consumer, but the advertising market's response to the consumer is delayed due to the limitations of prior art tools, most of which are unable to quickly and accurately integrate different sets of data. For example, television consumption data today exists independently of set-top box owner data and television original equipment manufacturers. Thus, the advertising strategy of television is planned according to the professional standards of television, and the network advertisement and the mobile advertisement including the subcategories such as social media are planned separately.

Furthermore, throughout the advertising industry, there are different entities for different media platforms (e.g., set-top boxes, telephones, and desktops). In different media, there are different data, data systems and data sources (suppliers). Today, these devices and media categories are still largely separate when incorporated into ad campaign strategies and programming.

Thus, in general, the business channel between the advertiser and the target audience is overly complex: the path from the generation of advertising content to the end consumer is tortuous, involving numerous participants-some of which are specific to certain media types, while others offer very specific and limited services in the supply chain. Advertisers prefer to be able to obtain inventory at as few points of sale as possible. Accordingly, larger media organizations, such as cable television companies, would benefit from the ability to obtain consumer data so that advertisers can be provided with a range of tailored advertising inventories.

Currently, some companies attempt to connect together a set of devices associated with a particular consumer, but they cannot handle different sources of such data at any reliable level of data integration or scale useful to advertisers. Determining the selection of a device by comparing and modeling incomplete user data with data of other similar users in a market segment is a partial solution to this problem, but existing approaches fail to create associations at a reliable or useful level of granularity.

Today, probabilistic and deterministic methods are not widely used to associate mobile and computer devices with an accurate audience or household. One of the reasons that this approach has not been adopted more widely is that data processing and pairing between different devices is inefficient. For example, for an association between a user and their respective devices of 1: predicting consumer's buying, viewing and advertising interaction habits at the 1 level, this approach is not sufficient to assume that any single device access instance represents the user's buying intent. This is due to modern media consumption habits-the user consumes on various devices and through different media (such as Hulu, Netflix or cable tv). Therefore, more sophisticated analysis is necessary to gain insight into the association of media consumption with the user's device group.

Another reason is that the consumer's buying habits cannot be measured more easily using probabilistic and deterministic methods, since accessing user device data is not easy to implement. For example, according to consumer privacy laws, it is illegal to access a user's device without the user's explicit consent. Thus, on a large scale, the combination of devices used by users and what media they consume on their respective devices is often unknown. This is a significant challenge for advertisers, as they need to determine which advertising inventory to purchase, and how to effectively reach the target audience best on a given category of devices.

Today, data systems that track consumer information for advertising targeting lack the ability to widely combine and integrate changeable and unchangeable consumer data categories (i.e., integration of multiple film layers is required). Most data systems contain a static, one-dimensional, homogeneous consumer classification. For example, two years ago a person 29 years old bought a car would be a consumer data point that would not be adjusted or updated. While it is easy to adjust the age of an individual over time, other changeable characteristics, such as wedding willingness, pregnancy, or other lifestyle changes, are not easily assessed or predicted.

Accordingly, there is a need for a method of integrating and linking given consumer data acquired over time from multiple different devices and using the aggregated data to reliably deliver advertising content on multiple devices. This would be beneficial if the degree of integration made the transaction between the advertiser and distributor simpler.

The discussion of the background art is included herein to explain the context of the technology. This is not to be taken as an admission that any of the material referred to was published, known or part of the common general knowledge as at the priority date of any of the claims appended hereto.

Throughout the description and claims of this application, the word "comprise", and variations of the word such as "comprises" and "comprising", is not intended to exclude other additives, components, integers or steps.

Disclosure of Invention

The present disclosure relates to consumer data and advertising inventory processing related to advertising content placement optimization between display devices including one or more televisions. In particular, the present disclosure includes methods of performing the same by a computer or network of computers. The disclosure also includes a computing device for performing the method, and a computer-readable medium having instructions for the method. The apparatus and methods of the present disclosure are particularly well suited for video content in online and television media.

In general, the present approach allows advertisers to assign media policies to different inventory types based on the probability of matching an audience category or type. In particular, the present technology relates to systems and methods for optimizing advertising campaigns, and contextual modeling as a way to increase revenue. The methods described herein improve the return on capital investment for advertising and improve the revenue for multi-channel video program distributors by improving efficiency and reducing the costs associated with determining advertising strategies.

In an alternative embodiment, the system develops an advertising strategy designed around the parameters of a given campaign (end-user). The strategies are directed to advertisements that occur in television, video-on-demand, display advertisements, and mobile and desktop environments.

The method includes analyzing consumer, media, and related data from an unlimited number of data inputs, including but not limited to: behavior (such as specific viewing and purchase history of individual consumers), and sources of demographics and location-related sources.

The present technology includes programmatically generating an appearance-like model in a consumer's device map to predict future consumption behavior. The method integrates the actual content consumption behavior with the broadcast user population and distributes the devices for consumption to individual consumers.

The present disclosure provides a method of distributing video advertising content to consumers on a television, the method comprising: receiving, from an advertiser, a price point and one or more campaign descriptions, wherein each campaign description comprises a schedule for placement of an advertising content item on one or more televisions available to a consumer and a target audience, wherein the target audience is defined by one or more demographic factors; determining one or more hard constraints associated with the one or more activity descriptions; defining a consumer pool based on a consumer attributes map, wherein the consumer attributes map contains information of two or more televisions and mobile devices used by each consumer, demographic and online behavior data of each consumer, and similarities between pairs of consumers, and wherein the consumer pool contains consumers with at least a threshold degree of similarity to target audience members; receiving an inventory list from one or more content providers, wherein the inventory list includes one or more televisions and online slots; determining one or more advertising targets, wherein each of the one or more advertising targets comprises a series of slots that are consistent with one or more campaign descriptions and one or more hard constraints and have a total cost that is consistent with the price point; performing an optimization of the distribution of the advertising content described by the one or more campaigns to one or more advertising targets based on one or more soft constraints, thereby producing one or more solutions; communicating a list of one or more solutions to an advertiser, wherein a solution comprises matching the campaign description to one or more slots in television content identified as likely to be viewed by a pool of consumers; and delivering the advertising content to consumers in the consumer pool through a first media channel on the television.

The present disclosure also provides a computer-readable medium encoding instructions for performing the methods described herein and for processing by one or more suitably configured computer processors.

The present disclosure also includes a computing device configured to execute instructions, such as instructions stored on a computer-readable medium for performing the methods described herein.

Drawings

FIG. 1 diagrammatically illustrates relationships between parties involved in advertising content delivery (e.g., advertisers, ad exchange platforms, media channels, and consumers);

FIG. 2 illustrates another set of relationships between parties using the techniques herein;

FIG. 3 shows a consumer diagram;

FIG. 4 illustrates a node in a consumer graph;

FIGS. 5 and 6 illustrate steps in creating a consumer graph;

FIG. 7 shows a flow chart of a method described herein;

FIG. 8 shows a flow chart of a method described herein;

FIG. 9 illustrates an apparatus for performing a method as described herein;

10A-10D illustrate aspects of an exemplary user interface.

Like reference symbols in the various drawings indicate like elements.

Detailed Description

The present technology is directed to computer-implemented methods that combine actual content consumption behavior, segments of the consuming population, and allocation of devices for media consumption to provide advertisers with a more targeted selection for advertising inventory purchases. These methods are particularly well suited for multiple multi-channel video program distributors (MVPDs) where the MVPDs have access to consumer data and can use these methods upon request by an advertiser.

Advertising functionality

The relationships between entities in the purchase, placement and consumption business of advertising content are depicted in fig. 1. As can be seen, the advertising ecosystem is complex, involving many different entities, as well as many different relationships. The methods and techniques herein may be used to simplify many of the relationships in fig. 1.

The advertiser 101 is a purchaser of the advertising inventory 109. An advertiser may be a company that has direct control over its advertising functions, or may be an agent that manages the advertising needs of one or more customers (typically corporate entities). The advertiser aims to have the advertising content 103 (also referred to herein as "advertisements") on each consumer's one or more devices 107 to act on one or more consumers 105 (typically a group).

For a given consumer, the device includes one or more of: televisions (including smart televisions), mobile devices (cell phones, smart phones, media players, tablets, laptops, and wearable devices), desktop computers, network photo frames, set-top boxes, game consoles, streaming media devices, and devices that are believed to be capable of operating within the "internet of things," such as household appliances (refrigerators, etc.), and other networked home monitoring devices, such as temperature controllers and alarm systems.

The advertising content 103 is typically created by the advertiser 101 or a third party with which the advertiser has signed up, and typically includes video, audio, and/or still images that seek to promote sales or to promote consumer awareness of a particular product or service. As further described herein, the advertising content 103 is typically delivered to the consumer through one or more intermediary parties.

Advertising content is typically of two different types: brand promotion and direct reaction marketing. The two types of time frames are different. The brand popularization and the understanding are improved; direct reaction marketing is intended to produce immediate reactions. For example, an automobile manufacturer may place directly reacting marketing materials into the marketplace and wish to measure the reaction of those who go to the dealer or website after seeing an advertisement. The methods described herein are applicable to two types of advertising content, but the measure of advertising effectiveness differs for these two types of methods: for example, the effectiveness of branding is measured by GRP (described further elsewhere herein), and the results of direct reaction marketing can be measured by website visitation.

When delivered to a mobile device such as a phone or tablet, the advertising content 103 may additionally or alternatively take the form of text/SMS messages, emails, or notifications such as alerts, banner pictures or badges. When delivered to a desktop or laptop computer or tablet computer, the advertising content 103 may be displayed as a pop-up within an application or browser window, or as a video that is designed to be played while other requested video content is being downloaded or buffered.

The consumers 105 are viewers and potential viewers of the advertising content 103 and may have previously purchased the product or service being advertised, and it is advantageous for the advertiser to have the viewer learn the product or service for the first time when they see the advertising content 103.

The advertising inventory 109 (which may also be inventory or available inventory herein) includes available slots or time slots 117 for advertising over a plurality of media interfaces or channels 111 through which the consumer may obtain information and advertising content. These media interfaces include television, radio, social media (e.g., LinkedIn, Twitter, Facebook, etc. online networks), digital billboards, mobile applications, and so forth. The media channels 111 may generate their own content 113 or may disseminate content from one or more other content providers or publishers 115. For example, a cable company is a media channel that delivers content from a multitude of television channel producers and content publishers. In general, media interfaces may also be referred to as content providers because they deliver media content 113 (television programs, movies, etc.) to consumers 105. One aspect of the techniques herein includes the ability to integrate inventory 109 from multiple media interfaces or content providers. The media channel 111 may also deliver advertising content 103 that has been purchased for delivery in time slots 117 to the consumer 105 for viewing on a variety of devices 107. The publisher 115 is typically a content owner (e.g., BBC, ESPN).

The slot 117 is a time, typically expressed as a time window (1 minute, 2 minutes, etc.) at a particular time of day (noon, 4:30 pm, etc.), or a window (e.g., 2-4 pm or 9 am to 12 am), or during a particular broadcast (e.g., a television program) on a particular broadcast channel (e.g., a television station or social media feed). The available slots are slots in inventory that advertisers purchase for placement of advertising content. Typically, another advertiser is available because it has not yet purchased it. As described further herein, a slot may also be defined by certain constraints, such as whether a particular type of advertising content 103 may be placed in a particular slot. For example, a sporting equipment manufacturer may have purchased a particular slot on a particular channel defined by a particular time of day, and may have purchased the right to deny other sporting equipment manufacturers the purchase of a slot within a certain time frame of the first manufacturer's slot on the same channel. In this context, a "hard constraint" is a legal or other mandatory limitation on ad placement in a particular time slot or specified media. "Soft constraints" refer to desired (non-mandatory) limits for placing advertisements in specific slots in specific media. "constraint satisfaction" refers to the process of finding a solution for a set of constraints that variables must satisfy. Thus, the solution is a set of values for the variables that satisfy all of the constraints.

In a broad sense, information refers to any content that a consumer can view, read, listen to, or a combination thereof, which may be used on a screen, such as a television screen, a computer screen or a display of a mobile device (e.g., a tablet, smartphone, or handheld/laptop computer), wearable (e.g., a watch, fitness monitor), display screen in a car or airplane, and the like. Information is provided by a media interface 111 (such as a television or radio station), a multi-channel video program distributor (such as a cable television provider, e.g., Comcast), or an online network (such as Yahoo | or Facebook).

VOD refers to a video-on-demand system that allows users to select, view, or listen to video or audio content of their own choosing, rather than having to view the content at a predetermined play time. Internet technology is often used to apply video-on-demand to televisions and personal computers. Television video-on-demand systems may transmit content via a set-top box, computer or other device, allowing for real-time viewing, or may download content to a device such as a computer, digital video recorder (also known as a personal video recorder), or portable media player for viewing at any time.

Communication between advertisers and media channels may be managed by a number of entities including: a demand provider (DSP)123, an ad exchange platform 119, and a supplier provider 121. The ad exchange 119 (also referred to herein as an exchange) is an environment in which advertisers can bid on available media inventory. The inventory may be digital, for example, by online placement on the Internet or by digital radio such as SiriusXM, or analog, for example, by television channel such as ESPN, CNN, Fox, or BBC, or FM/AM radio. The ad exchange 119 is typically specialized to handle certain types of content. For example, SPocX is specific to digital content and WideOrbit is specific to programming television.

The supplier provider (SSP)121 is an intermediary that takes inventory 109 from the media channel 111 and optionally may provide the inventory to the supplier provider (DSP)123 through the ad exchange platform 119 so that advertisers may purchase or bid on the inventory when deciding how to locate the advertising content 103. SSPs are sometimes classified into public or private categories depending on whether the media channel can limit the identity and number of advertisers that obtain inventory. In some cases, if the functions of the ad exchange platform upon which the purchaser of the ad content relies are performed by one or both of the DSP and the SSP, the SSP interacts directly with the DSP without the need for the ad exchange platform. The techniques herein are particularly suited to implementation and execution by a suitably configured DSP.

In one configuration, the ad exchange platform 119 interfaces between a supplier provider (SSP)121 and a demand provider (DSP) 123. The splicing action includes receiving inventory 109 from one or more SSPs 121 and providing it to the DSP, and then receiving bids 125 for the inventory from the DSP and providing these bids 125 to the SSP. Thus, the DSP enables advertisers to bid on inventory provided by a particular SSP (e.g., SPotX or WideOrbit). In some configurations, the DSP takes much or all of the role of the ad exchange platform.

An advertising campaign (or campaign) is a plan for advertisers to deliver advertising content to a particular consumer population. A campaign will typically include a selection of advertising content (e.g., a particular advertisement or multiple forms of advertisements, or a series of related advertisements intended to be viewed in a particular order), and a time at which the campaign is to be conducted (e.g., 1 week, 1 month, 3 months). The advertiser typically sends the campaign description 127 to the ad exchange platform 119 or DSP121 and receives as feedback a list of available inventory 109. The campaign description 127 includes one item of advertising content 103 and one or more types of targeted devices 107, or a schedule for delivering two or more items of advertising content 103 in succession between one or more devices 107. Campaign description 127 also includes a description of the target audience, where the target audience is defined by one or more demographic factors selected from, but not limited to, age range, gender, income, and location.

The DSP123 then provides an interface through which the advertiser 101 can associate the campaign description 127 with the inventory 109 and make purchases or bids for a plurality of slots 117 in the inventory. The DSP123 or the transaction platform 119 may provide sets of inventory that match a given activity description 127: each set of library occurrences that match a given campaign description is referred to herein as an ad target 129. The advertiser 101 may select the target or targets from the list of advertising targets that it wishes to purchase. Once it has purchased a particular target, the SSP121 is notified and delivery instructions 137 are sent to the various media channels 111 so that the advertising content 103 or selected content 113 can be delivered to the relevant consumer in the applicable slot 117.

The purchase of a given slot is not simply a sale at a given price, but rather is accomplished through a bidding process. The DSP will bid on multiple slots and determine a bid price to submit to the SSP for each slot. For a successful bid, the SSP delivers the advertising content to the media channel and ultimately to the consumer. Bids placed on a particular goal are generally higher than bids placed on all goals.

The bidding process depends in part on the type of advertising content. Television content can be scheduled in advance, while for online content, the usual bidding structure is "just-in-time" bidding: advertisements are only delivered when a particular consumer is visible online. In general, the methods herein are independent of the bidding process and are applicable to any commonly employed bidding method, including real-time bidding, and bidding using program television data details.

Tags are provided to a given online advertisement using protocols such as VPAID (https:/en.wikipara.org/wiki/Mipo) or VAST (video advertisement service template), and tag collections include data about whether a consumer clicks on or views content. Tags typically contain a plurality of data items of how consumers interact with advertising content. These data items may be returned to the SSP and/or DSP to provide feedback regarding the placement of the advertisement. For example, the data items may include data related to whether the user clicked on the video online. Some data items correspond to events known in the industry as "beacon" events because of their importance to the advertiser: for example, a beacon event may include the fact that a user stopped a video clip before the video was completed.

The process of generating advertising targets may also depend on one or more campaign requirements. Campaign requirements, as used herein, refer to financial constraints such as budgets set by advertisers or other purchasers of advertising inventory, and behavioral specifications such as goals for multiple consumers. The campaign demand information is used with the campaign description when making purchases or bids on inventory.

The multiple DSPs 123 also provide the advertiser 101 with aggregated consumer data and device data from different sources. These data help advertisers select the inventory, time slots, and media channels that best suit their goals.

The data used by the multiple DSPs may include census data 131, or data about a particular consumer and device 133. The census data 131 includes demographic data that may be used to optimize inventory purchases. Thus, the census data 131 may include demographic data, such as age distribution of the population, income differences, and marital status in a particular viewing area, which is independent of the media interface actually viewed by the demographic members. The census data 131 may be aggregated from a variety of sources, such as state and county records, and census bureau of america data.

The Data Management Platform (DMP)135 may provide other types of third party data 133 to the DSP regarding the consumer and the device it uses. Typically, DMPs provide data storage facilities with embedded functionality. The DMP may download data and perform various analysis functions including sorting, storing, processing, applying matching algorithms, and providing data output to buyers and users. Examples of DMP's include Krux, Exelate, Nielsen, Lotame. The consumer and device data 133 that can be delivered to the DSP from the third party provider can supplement other consumer and device data 143 provided by the media channel. Data on consumers and their devices used are advertiser-related, including viewing habits and specific behavioral data that can be retrieved directly from the media channel. For example, as discussed further elsewhere herein, when a media channel provides an advertisement to a consumer, the channel may collect information about the manner in which the user accessed the advertisement. Due to the enormous amount of data involved, a media channel may also fail to provide any information about a particular consumer after a relatively short period of time (e.g., 14 days). In this case, the DSP may obtain the user's data from a third party (e.g., DMP). The third party may also obtain the data offline. Offline events, as used herein, refer to events that occur independently of internet or television viewing: for example, it may include purchases from stores and other types of location-based events that advertisers may consider important. Data can be shared and transferred between entities herein (e.g., between a DMP and a DSP, between a DSP and an SSP, and between a media channel and an SSP or ad exchange platform) using any commonly accepted file format: these formats include, but are not limited to: JSON, CSV, and swift, and any appropriately formatted text file format.

Role of video distributor of multi-channel program

Fig. 2 shows a set of optional relationships between different entities, where a multi-channel video program distributor (MVPD) plays a major role.

In fig. 2, advertisers and agents interact directly with MVPD211 as a user 101 through placement of a campaign description 127, advertising content 103, and bidding 125 on inventory 109. The MVPD has provided their inventory 109 (typically the inventory 109 is television content 113, but may also include online data from publishers 115) to advertisers so that the advertisers may bid on the inventory 109. The MVPD collects a database of consumer and device data 143 (e.g., television viewing habits) and tools for analyzing it. As described elsewhere herein, MVPD assists advertisers in determining the most appropriate slot in inventory and then ensures that advertising content 103 is delivered to a consumer target group.

Define a limit

Exposure refers to any instance of an advertisement arriving at a consumer. On a television, assuming the television is playing an advertisement, the owner of the television or the general viewer will see the advertisement and the show counts as one exposure. If there are multiple people in the same household, the number of exposures may be equal to the number of people who can watch television. In an online environment, exposure occurs if a consumer is viewing a web page and an advertisement is displayed on the web page in a pop-up form, or the user clicks on a link that causes the advertisement to run.

An audience segment is a list of consumers who use cookie synchronization or other methods to identify from their personally identifiable information that a consumer is of a certain type or is associated with an activity (purchase, television viewing, website access, etc.).

As used herein, online means connected to the internet or another computer network (e.g., an intranet) that allows multiple devices to communicate with each other. A device is online if it accesses the internet or other network through a WiFi connection or through a cellular data network, or by using a short range communication protocol such as bluetooth. Thus, online devices generally include, but are not limited to: computers, such as personal or desktop computers, workstations, laptops, notebooks, and tablets; electronic books (e.g., Nook); mobile devices (such as tablet computers (apple ipad, samsung Galaxy, etc.) and mobile phones or network music players, mobile hotspots, items displayed on vehicles (such as cars, buses, trains, buses, and airplanes), wearable devices such as watches, fitness monitors, virtual reality viewing equipment (such as Oculus), and devices belonging to the general "internet of things" category, such as home appliances, including but not limited to smart televisions, refrigerators, digital photo frames, thermostats, and security systems.

Cookie synchronization refers to a process that allows data exchange between a DMP, an SSP, and a DSP, and more generally, between a content publisher and an advertisement purchaser. cookies are files in a mobile device or desktop computer that are used to save and restore information about a particular user or device. The information in the Cookie is typically protected so that only the entity that created the Cookie can later retrieve the information from the Cookie. Cookie synchronization is one way in which one entity can obtain information about a consumer from a Cookie created by another entity without having to obtain accurate identification of the consumer. Thus, given information received from a media channel about a particular consumer, through Cookie synchronization, further information about that consumer can be added from the DMP.

For mobile devices, one device ID corresponds to a unique specific device. For television there is a hashed IP address. The device ID information may be used to link a group of devices to a particular consumer, as well as to link multiple consumers (e.g., in a given household) to a particular device. The DSP may collect data stores associated with mobile device IDs and television addresses that add "cookie" data over time.

Cross-screen refers to media data (including advertising content) distributed among multiple devices of a particular consumer, including: for example, a television screen, a computer screen or a display screen of a mobile device (such as a tablet, smartphone or handheld/laptop computer), wearable (such as a smartwatch or fitness monitor), in-car or in-plane display screen, or a display on a networked appliance (such as a refrigerator).

The arrival rate refers to the total number of different people who have viewed an advertisement at least once during a period of time.

In a cross-screen advertising or media campaign, the same consumer may be exposed to advertisements multiple times using different devices (e.g., televisions, desktop or mobile devices). The repeat arrival rate refers to the number of different people who see the advertisement regardless of the device. For example, if a particular consumer sees an advertisement on his/her television, desktop, and one or more mobile devices, the consumer only contributes 1 to the arrival rate.

The increased arrival rate is an additional repeat arrival rate for an activity that exceeds and exceeds the arrival rate obtained prior to starting the activity (e.g., obtained at a previous activity). In one embodiment herein, an activity type may include a television expansion: in this case, the advertiser has initiated an advertising campaign on the television, but the revenue is declining. Advertisers want to improve campaigns for the digital marketplace in order to increase reach. In this way, the DSP can inherit activities that have been conducted on one or more media channels.

In addition to television programming content and online content delivered to desktop and mobile devices, advertisements may also be delivered in OTT content. OTT (from the term "over the top") refers to the delivery of audio and video over the internet without MVPD participating in controlling or distributing the content. OTT content is thus any content that is not related to a particular box or device. For example, Netflix or HBO-Go may deliver OTT content because the consumer does not need a specific device to view the content. In contrast, MVPD content, such as that delivered to a cable or set-top box, is controlled by a cable network or satellite provider (e.g., Comcast, AT & T, or DirecTV) and is not OTT content. Specifically, OTT refers to content from third parties (e.g., Sling TV, YuppTV, Amazon instant video, Mobibase, dramantze, presto, DramaFever, Clash, HBO, Hulu, MyTV, Netflix, Now TV, Qello, RPI TV, Viewster, where everkv, Crunchyroll, or WWE networks) and is delivered to end-user equipment, leaving an Internet Service Provider (ISP) to assume only the role of transmitting IP packets.

Furthermore, an OTT device is any device that is connected to the internet and that is capable of accessing a variety of content. For example, Xbox, Roku, TiVo, Hulu (and other devices that may run on a cable television network), desktop computers, and smart televisions are all examples of OTT devices.

The gross-reception-point (GRP) refers to the size of the advertising campaign in terms of schedule and media channel involved, and is given by the number of exposures per target audience, expressed as a percentage (hence, GRP may be a number > 100). For example, if an advertisement is up to 4 times 30% of the los Angeles population, the GRP is 120. (data can be measured, for example, in a Nielsen panel consisting of 1000 viewers at l.a.).

The Target Reception Point (TRP) refers to the number of exposures per target audience member based on the sample population. This number relates to the individual: for example, in l.a., advertisers want to target males aged 25 and older. If there are 100 such people in los Angeles, and 70% see the advertisement, then TRP is 70% x the number viewed.

"Cross-screen" refers to the analysis of media, consumer, and device data that incorporates audience data for multiple devices.

"high frequency" refers to high frequency transactions related to ad buying and selling. The methods and techniques herein may be implemented by an ad exchange platform that utilizes a computer to process bid requests for large volumes of ad inventory at high speed. The system herein may operate at a high frequency, for example 10000 to 100000 viewer exposure queries per second. The query may be dynamic and real-time.

Consumer data

Data about consumers can be divided into two categories: one is an immutable feature, such as race and gender; another category is changeable characteristics such as age, occupation, address, marital status, income, taste and preferences. Where multiple changeable characteristics, such as occupation, may change over time, while others, such as age, change at a consistent rate. Today, data systems for tracking consumer information to target advertising content lack the ability to track both types of consumer data extensively. Most data systems contain a static, homogenous classification of consumers. For example, a 29 year old person who buys a car two years ago is a consumer data point that does not update or increase over time. Even though the age of the individuals stored in the system may be adjusted over time, other changeable characteristics, such as changes in marital status or lifestyle changes, are not taken into account in this consumer classification.

At various stages of the methods herein, it is described that each consumer in a consumer population is treated in a particular way by: for example, the computer may be programmed to analyze the data for each consumer in its database to determine which, if any, watched a particular television program or visited a particular website; alternatively, some comparative analysis may be performed in which attributes of each user in one category of population are compared to attributes of each consumer in another category of population. Each group of people may include thousands, or hundreds of thousands, or even millions of people. It is assumed that these methods, when applied on suitable computing resources, are capable of performing certain calculations and operations on each member of the population. However, it is also consistent with the method herein, i.e., "per consumer" in a population may also mean the majority of consumers in the population, or all consumers in the population that make the calculation method feasible. For example, one or more given consumers in a population are omitted from a particular calculation because there is not enough data on the individual, and this does not mean that the population is not sufficient for analysis to provide a meaningful result. Thus, when referring to a population of potentially millions of consumers, "each" does not necessarily mean every member of the population, but may mean a large and virtually reasonable number of members of the population, in order for a given calculation to yield a result.

Consumer graph

Each node in the consumer graph represents a consumer (or a single user). The technique represents multiple implementations with weighted graphs, where relationships between consumers (nodes) are defined as similarities (boundaries). Consumer graphs are used herein to categorize, store and integrate large amounts of consumer data and allow entities such as DSPs to establish connections between the data used to construct the consumer graph and other data (e.g., television viewing data) through the data on a given consumer device.

One way to construct the graph is to use deterministic relationship data; another approach is to use the attributes of each node. In some cases, a combination of deterministic and probabilistic approaches may be employed. In a relatively simple deterministic approach, the basis is to have the exact data of the consumer, such as login information from the publisher. Thus, if a person logs on to different devices multiple times with the same ID, it can be ensured that the identities of the persons are matched. However, this exact information may not always be available. In contrast, in a probabilistic approach, it is necessary to make inferences: for example, if the same device is seen in the same location, or similar behavior occurs on a given device at different times, it can be inferred that the devices belong to the same user.

In some embodiments herein, machine learning methods are used, as well as Bayesian (Bayesian) and regression algorithms to explore commonalities between consumers. These methods are useful in cases where the parameters to be considered are of limited number. In some other embodiments, deep learning techniques are more useful in finding consumer likeness points and building consumer graphs. Machine learning is the preferred technique to match the exact pieces of information, e.g., whether two consumers have visited the same web site. Deep learning may explore the details of a particular video or television program-e.g., by analyzing natural scene statistics-to determine whether, for example, two advertisements viewed by a particular consumer have in common, in addition to subject matter. For example, two advertisements may contain the same actor, although the depicted products have little in common, and even though the depicted products have little in common, may be preferred by consumers.

In a preferred embodiment, the device graph described herein is based on probabilistic data. The probabilistic approach to graph building uses behavioral data (e.g., viewing habits) to match users.

In some embodiments, an entity such as a DSP may construct a device map; in other embodiments, it may be obtained (e.g., purchased) from another entity, such as a DMP.

In various embodiments herein, both the device map and the consumer map operate together in a manner that allows mobile data to be bound to television data.

The term graph is used herein in its mathematical sense as a set G (N, E) of nodes (N) and edges (E) connecting pairs of nodes. Graph G represents the relationship between nodes: according to certain criteria, two nodes connected by an edge are similar to each other, and the weight of the edge defines the degree of similarity. Pairs of nodes that do not meet similar criteria cannot join through edges. FIG. 3 illustrates the concept of a graph showing 6 nodes, N₁-N₆Wherein the three pairs of nodes are connected by edges.

In the embodiment of the figures herein, node N is an entity or object having a set of attributes A. In fig. 3, each node has an attribute array associated with it, denoted as Ai for node Ni.

In the embodiments of the figures herein, the presence of an edge E between two nodes indicates that there is a relationship or degree of similarity between the two nodes above a defined threshold. The weight w _ E of an edge is the similarity of two nodes. The weights of the edges in FIG. 3 are shown in thickness (in this case, w _ E)₁₂>w_E₃₄>w_E₁₅)。

In the consumer graph, a node represents one individual or a family consisting of two or more individuals, and has a series of attributes such as the sex and age of the individual, the history of watching television programs, websites visited, and the like.

FIG. 4 illustrates an exemplary structure of a node of a consumption graph. Each node has a collection of attributes, including type and behavior, whose data is continuously collected from first and third party sources. Many attributes may be changeable if new information is available from the consumer, and the set of attributes (i.e., the number of different attributes stored for a given consumer) may also increase over time as new data about the consumer is collected. One aspect of the present technology is that the graph is built from potentially unlimited number of inputs for a given user, such as online, offline, behavioral, and demographic data. These inputs will update over time and allow for refinement of data for a particular consumer and for data to be available for expansion across a population of consumers. The type and nature of the data that can be used is not limiting, meaning that the approach here is superior to that employed by advertising companies that rely on static data sets and fixed populations.

Some sources of collected data are as follows.

The type data is classification data about consumers that are not generally changed, that is, are immutable. The behavioral data is continuously updated based on the consumer's recent activity.

Each node comprises a combination of one or more devices (desktop, mobile, tablet, smart tv). For each device, data based on the user type of the device is collected from third parties and first party sources.

Table 1 shows an example of data sorted by category and source.

TABLE 1

The data of the first party comprises data about user behavior, such as: purchases, viewership, site visits, etc., as well as revenue, gender, etc. types of data provided directly by publishers for better locating and reporting their own activities. (e.g., a coca-cola company may provide a list of users to the DSP that "like" coca-cola products on social media to improve their video advertising campaign.) the first party type data may be collected from advertisements played directly on the device and information collected from the device (e.g., one or more IP addresses). The first party type data includes a location from the IP address, a geographic location of the mobile device, and whether the device is located in a business or residential property.

The third party type data may be obtained from an external supplier. Information about cookies or devices is provided by an external vendor, such as a DMP, such as Krux (http:/www.krux.com /), Experian (providing purchase behavior data), or Adobe, for one-to-one cookie synchronization or device synchronization. Example data includes market share of the consumer, preferences such as age range, gender, income level, education level, political affiliation, and brands that are liked or attended to on the consumer's social media. In addition, the external supplier may provide type data based on the recent purchase of the device. Third party data includes information such as gender and income, as such data is not collected directly from external suppliers. Third party data can be collected without providing advertising services. The television programs viewed and the purchases are third party data.

The first party data is typically generated by the DSP; for example, the first data is data that the DSP may collect from a service advertisement or a brand/organization providing the data. The first party data includes data that is dependent on the advertising service to access it.

Behavioral data may be collected from the device by the first party and third party sources. The behavior is typically first party data, which is variable.

The first party behavioural data is collected from advertisements provided directly to the device. This includes websites visited, television programs or Ott viewed through the device, or Video On Demand (VOD) content.

Third party behavioural data is obtained from external suppliers, typically DMP's such as Experian, Krux, adobe, Nielsen and comScore, and ad exchange platforms or networks such as Brightroll, SPocx, FreeWheel, Hulu. Example data includes a history of television programs watched on the device in the last month, a history of websites visited on a personal computer or laptop or mobile device, and a history of event-based locations in the mobile device (e.g., whether the device is in starbucks). In some cases, the same type of data may be obtained from the first party entity and the third party entity.

Edges between nodes in the consumer graph indicate that consumers have a threshold degree of similarity, or interact with each other. For example, if the nodes are physically close, or computed probabilistically based on similarity of attributes, the edges may be computed deterministically. Probabilistic methods used include, but are not limited to: k-means clustering and connected domain analysis (based on graph traversal methods, including constructing a path from one vertex to another). Since attributes are changeable, if the similarity score of a pair of nodes changes, the edges change, whether in their weight or by creating or canceling the attributes. Thus, the graphics are not static and may change over time. In some embodiments, the change is dynamic: the similarity score is continuously recalculated as the node attributes are updated.

Typically, attributes and data are added dynamically (as they are acquired). The graph may be reconstructed weekly to take into account new attributes and data, to establish new weights for the edges, and to identify newly connected or reconnected devices. (the graph construction and reconstruction can be done in the cloud, i.e. under DSP control by distributing the work over multiple processors on a computer network or over processors in a data center.)

Calculating a similarity S between the two nodes N _1, N _2 according to the similarity measure, the similarity S being the reciprocal of a distance function f (N _1, N _2), f (N _1, N _ 2): n _1, N _2- > s, which defines the similarity between two nodes according to their properties.

In the consumer graph, the similarity represents the similarity of the demographic attributes and the observation preferences of two people. Similarity may be computed by attribute, and then individual similar attributes are weighted and combined together to generate an overall similarity score for a pair of nodes.

When the attributes of two nodes are represented by binary vectors, a multiple-time metric may be used to define the similarity between a pair of nodes based on the attributes. Any of these metrics are suitable for use with the techniques herein. In some embodiments, to improve storage efficiency, a binary vector may be represented as a string or array of strings of bits.

When using the similarity measure of the reciprocal of the distance function f (N _ i, N _ j), a zero value of the distance function indicates that the type and behavior of the two nodes are the same. Conversely, a larger value of the distance function indicates that the two nodes are different. An example of a distance function is the euclidean distance,

f(N_i，N_j)＝||A_i–A_j||^2

where A _ i and A _ j are sparse vectors representing the properties of nodes N _ i and N _ j, and the distance is calculated as the sum of the squares of the difference values of the corresponding components of each vector.

The comparison between binary vectors or between bit strings may be done according to one or more of several similarity measures, the most common of which is the tamiouto coefficient. Other common metrics include, but are not limited to: cosine, Dice, Euclidean, Manhattan, city block, Euclidean, Hamming, and Tverseky. Another distance metric that may be used is LDA (latent Dirichlet allocation). Another way of defining the distance comparison is by deep learning embedding, by which the best form of the distance measure can be learned instead of fixing it in the form of e.g. a cosine distance. One example of a method is through manifold learning.

The cosine dot product is a preferred metric that can be used to define the similarity between two nodes in the consumer graph. The cosine similarity, i.e. the dot product of A _ i and A _ j, is defined as follows:

f(N_i,N_j)＝A_i.A_j

in this case, each vector is normalized so that their size is 1.0. The cosine similarity measure has a value of 1.0, indicating that the two nodes are identical. Conversely, the closer the value of the cosine metric is to 0.0, the more dissimilar the two nodes are. The cosine measure can be converted to a distance-like quantity by subtracting its value from 1.0:

f'(N_i,N_j)＝1–A_i.A_j

an example of a more complex distance function is a parameterized Kernel, such as a radial basis function.

f(N_i,N_j)＝exp(||A_i–A_j||^2/s^2)，

Where s is a parameter.

In a more general case, a bit string is a vector containing numbers other than 1 and 0 (e.g., it contains percentage or non-normalized data), and then the similarity can be calculated from a distance metric between the number vectors. Other metrics, such as Mahalanobis distance, may also be applicable.

Typically, the similarity score S is a number between 0 and 100, other normalization methods may be used, such as a number between 0 and 1.0, a number between 0 and 10, or a number between 0 and 1000. The scoring system may also be non-standardized and simply expressed as a number that calculates a ratio of similarity between two consumers.

In some embodiments, in calculating the similarity score, each contribution factor may be weighted according to a coefficient that enables the relative importance of the factor. For example, a person's gender may be weighted more heavily than whether they are watching a particular television program. The weights may be initially determined by applying heuristics and may ultimately be derived by statistical analysis of advertisement campaigns that are continually updated over time. Other methods of deriving weighting coefficients for determining the contribution of an attribute to a similarity score include: regression, or feature selection, such as least absolute shrinkage and selection operator ("LASSO"). Alternatively, it may be adapted to "real data", such as login data. In some embodiments, when the system tries different combinations or features, the system can infer a higher precision/recall by using a "hold out" test data set (where the feature is not used for the construction of the graph).

Another way to derive a similarity score for a feature is to analyze data for a continuous comparison from an ad campaign to consumer feedback using either: machine learning; neural networks and other multi-layer perceptrons; a support vector machine; analyzing a main component; a Bayesian classifier; fisher judging; linear discrimination; maximum likelihood estimation; least square estimation; logistic regression; a Gaussian mixture model; a genetic algorithm; simulated annealing; a decision tree; projecting likelihood; k-nearest neighbor algorithm; function discrimination analysis; rule integration prediction learning; processing a natural language; a state machine; a rule system; a probabilistic model; expectation maximization; a maximum entropy markov model. Each of these methods may evaluate whether a particular attribute of the consumer is appropriate for measuring the effectiveness of an advertising campaign and provide a quantitative weight for each attribute.

Mode of presentation

To properly evaluate the entire consumer population, a large number of nodes need to be stored. Furthermore, the set of attributes that represent node type and behavior can be quite large. Storing a large set of attributes for these nodes is challenging because the number of nodes can be as high as hundreds of millions. Efficient storage of data is also important because graph computations can be done most quickly and efficiently if node data is stored in memory.

In a preferred embodiment, the attributes are represented by sparse vectors. To accomplish this representation, a collection of all possible node attributes of a given type is stored in a dictionary. The type or behavior of each node is then represented as a binary sparse vector, where 1 and 0 represent the presence and absence of an attribute, respectively. Since the number of possible attributes of a given type is very large, most entries will be 0 for a given consumer. Thus, only addresses of those non-zero attributes need to be stored, and each sparse vector can be efficiently stored, typically with all vectors occupying less than 1/100 of the space.

For example, having these attributes encode the television programs that a given consumer has watched in the past month. The system lists all possible tv programs in the dictionary, which may have up to 100,000 different programs. For each node, the user has viewed the program in the last month as a 1, otherwise it is a 0.

If the attribute is to indicate a different revenue level, then multiple revenue levels are enumerated, with 1 indicating that the consumer belongs to a particular revenue level (and all other entries are 0).

Thus, for a consumer, i, who has an annual income between $30,000 and $60,000 and has seen "mad car show in the uk (Top Gear)" in the last month, the following is established:

TV _ dictionary { "Walking meat (Walking Dead)", "Game of powers (Game of threads)", … …, "Top Gear (uk mad car show)" }

TV_i＝[0,0,…,1]

TV _ i can be simply stored as [4 ]; only the fourth element of the vector is non-zero. Also, in terms of revenue:

income _ dictionary { < $30,000, $30,000- $60,000, $60,000- $100,000, > $100,000}

Income _ i is [0,1,0,0]

Revenue _ i can simply be stored as [2] because only the second element of the vector is non-zero.

Therefore, all the attributes of node i can be efficiently represented by a sparse vector. This requires 2 to 3 orders of magnitude less memory than dense representation.

Graph construction

FIGS. 5 and 6 show a flow chart of steps for building a consumer graph. MVPDs may use their data alone to build graphics, or may obtain consumer graphics from third parties (e.g., DSPs), or they may build graphics based on a combination of data from both sources.

Initially, the graph is a collection of devices that are mapped to consumers. Multiple data sources are used to group multiple devices (tablet, mobile, television, etc.) onto a single consumer. This is typically done using polymerization techniques. In order to home a single device (e.g., a smart television) to multiple consumers, segmentation techniques are used.

Using the aggregation approach, multiple devices may be grouped into a single consumer (or graph node). Some data sources used for this purpose include, but are not limited to:

IP address: a plurality of devices belonging to the same IP address represent a single user or home.

Geographic location: using latitude and longitude, multiple devices in the vicinity of the location may be attributed to a single user.

The publisher logs in: if the same consumer is logged in from multiple devices, the devices may be associated with the consumer.

In this process, the identity of the consumer is masked to eliminate privacy concerns. The result is that a single customer ID is linked with a particular device.

Let P (d _ i, d _ j) be the probability that two devices d _ i and d _ j belong to the same node (consumer or home). From multiple data sets obtained from different classes of devices, such a probabilistic expression can be constructed:

P(d_i,d_j)＝

w_IP×P(d_i,d_j|IP)×w_Geo×P(d_i,d_j|Geo)×w_Login×P(d_i,d_j|Login)/Z

where "x" means "multiplication", where w _ is a weighting factor and P (d _ i, d _ j | Y) is a conditional probability (the probability that device i and device j belong to the same user if Y has the same value for both devices and Z is a standard factor). Thus, Y may be an IP address. (the value of the conditional probability may be 0.80). Each data source has a different weighting factor: for example, the login data may be weighted higher than the IP address. The weights may be fixed or learned from a separate set of verification data.

Once multiple devices are grouped into a single node, the types and behaviors from the individual devices are aggregated into attributes for the single node. For example, attributes (and corresponding sparse vectors) from movement (e.g., location events) and desktop (recently purchased) are aggregated. This provides the user with more comprehensive information, allowing for more accurate and meaningful inferences to be made about the node.

Associating devices with a given consumer is made possible by the data associated with these devices and known to various media channels. For example, smart televisions store location information and subscription information related to the content they broadcast. This information is shared with and available from other entities, such as cable company companies. Similarly, a mobile device (such as a tablet or smartphone) may be associated with the same wifi network as the (in-home) smart tv. Thus, information about the location may be shared with, for example, the cell phone operator and the broadcaster of the mobile device that subscribes to the content. One key aspect of the graphical approach herein is that it allows consumer information to be connected across different devices and media platforms (which are typically separate from each other): in particular, the graphics herein enable linking consumer data from online and offline purchase and viewing sources with television viewing data.

By the subdivision approach, for example, a single device (e.g., a smart television) may be associated with multiple consumers (or graph nodes), e.g., they have mobile devices connected to the same wifi network as the smart television.

Given a node n to which multiple devices are assigned, these multiple attributes are aggregated into smaller groups of devices, e.g., one TV ID connected to multiple devices from a public IP address. Tv ratings data is a summary of the attributes of all devices. Clustering algorithms (e.g., k-means clustering) may be used to group devices into smaller clusters. The number of clusters k may be set according to the number of devices (by default, k ═ number of devices/4). Sometimes, only the comprehensive data at the family level may be collected. For example, there may be up to 20 devices in a home. However, by using the behavior data, it can be determined that there are 4 main clusters of these 20 devices, e.g., 5 devices as one cluster, where the clusters correspond to different individuals in the same household. Thus, despite the two types of devices (shared devices and personal devices), it is still important to attribute behavioral data to the user.

Once a shared device is homed to multiple nodes, the data collected from the device may be homed to the nodes. For example, television viewing data for smart televisions may be collected from OEMs. By this attribution method, television viewing data can be added to the set of attributes of the node. Finally, the smart tv may belong to different people of the same household.

Approximate modeling by learning a distance function

Given a picture G (N, E) and a functional form defining a similarity measure, and a set of seed nodes, a set of "approximate" nodes similar to the seed nodes may be generated, where the similarity is a fixed function or defined by a learned function. This is useful when identifying new consumers who may be interested in the same or similar content as a group of consumers known to the advertiser. Similar principles may be used in predicting likely behavior of consumers from historical data of consumer groups.

The seed node may be a set of nodes, such as a home node or an individual node, from which a set of similar nodes is generated using a fixed or learned similarity metric. For example, a seed node may be defined as an audience section (e.g., a list of users watching a particular program). This is useful for determining the members of each audience segment, which may have similar viewing habits even though they do not watch the exact same programs as those watched by the seeds.

Given a set of seed nodes (and their attributes) in the graph, the output of the similarity modeling is a set of nodes (including the seed nodes) that are similar to those nodes based on fixed or learned similarity metrics.

Several different vectors can be used to determine the approximate model: one is the vector of the television program. This vector may be up to 40k elements long. Another vector is a list of consumers who watched a particular program (e.g., Simpsons-The Simpsons). The audience vector for a given television program may be up to 10M elements in that it contains one element for each consumer. Another vector would be the vector of visited web sites (e.g., 100k elements in length). Yet another vector would be based on the online video being viewed (up to 100k elements in length).

Generally, the tv program comparison data has access to 10M user groups. The online data may identify a potentially larger audience, such as 1.5 million consumers. It should be understood that television data may be accumulated on a variety of television consumption devices including, but not limited to, linear, time-shifted, conventional, and programmed devices.

The similarity between two different nodes can be computed from their properties represented by sparse vectors. Given a distance function f (N _ i, N _ j) and a set of seed nodes N _ S, the pair-wise distances between each element of the seed node (N in N _ S) and all other nodes N' except the seed node N are calculated. That is, all f (n, n') are calculated.

After calculating all node pair similarities, only nodes meeting f (n, n') < T are selected. T is the maximum distance threshold within which the nodes are considered similar. Additionally, the values of f (n, n ') (where n is not n') are arranged in decreasing order and the top t node pairs are selected. In either case, T and T are parameters that are preset (provided to the method) or parameters that are learned from the real data or the validation data. The set of all nodes n' satisfying the above condition constitutes a set of "approximate nodes".

Graphical reasoning

Given graph G (N, E), the possible attributes of node N may also be inferred from the attributes of the nodes adjacent to those nodes in the graph. This approach may be useful when incomplete information exists for a given consumer, but sufficient information can be inferred therefrom. For example, television rating attributes may be missing for node n (typically, real information may be obtained if the user does watch the program, or does not know whether they watched the program), however those attributes may be obtained from the neighboring nodes n', n "in the graph. Nodes n, n' and n "contain all other attributes such as the level of income and web sites visited.

In another example, it is useful to calculate the probability that a consumer associated with node n is watching a "Walking meat" program, because both n ', n' are watching "Walking meat". Given the weight of the edges between n and n ', n ", the similarity is w' 0.8 and w" 0.9, respectively, and the likelihood of watching a program n is 0.9 based on his/her own attributes, then this probability is given by the following equation:

p (n view "walking meat)")

＝[0.8×0.9+0.9×0.9]/[0.8×0.9+0.9×0.9+(1–0.8×0.9)+(1–0.9×0.9)]

＝0.765

Similar principles may be used in predicting the likely viewing behavior of consumers from historical data of consumer groups.

Rate of accuracy

The graph will continually improve as new data is received. In one embodiment, techniques such as machine learning are used to improve the quality of the graphics over time. This may be done periodically, for example in a weekly build phase. Consistent with the approach herein, the graph used is updated frequently when new consumer data is available.

To determine the accuracy of the graph, precision and recall may be compared to the validation data set. The verification dataset is typically a (sub-) graph where the device and node relationships are explicitly known. For example, login information from an online network (e.g., eHarmony) indicates when the same user logs into the site from different desktop (Office, laptop) and mobile devices (smart phone and tablet). Thus, all devices that are often used to log on to a site are bound to the same consumer and thus to the individual's graph node. This information can be used to verify that the constructed graph binds these devices to the same node.

If D is the set of devices in the verification set, Z (D) represents a graph containing a set of nodes constructed by the set of devices D. For different data sets, and different graph construction methods, it is possible to obtain different z (d).

For set Z (D), the probability of True Positive (TP), False Positive (FP) and False Negative (FN) can be calculated. True positive means that all nodes in z (d) are also nodes in the verification set. False positive (false positive) means that all nodes in n (d) do not belong to nodes in the verification set. False negative (false negative) indicates that all nodes belong to the verification set but not to Z (D).

Precision, defined as TP/(TP + FP), is the proportion of retrieved devices that are correctly grouped to consumer nodes.

The recall, defined as TP/(TP + FN), is the proportion of consumer nodes that are correctly grouped.

There are different tradeoffs between precision and recall depending on the application used. In the case of building a consumer graph, it is preferable to obtain a high precision and a high recall to be able to be used for comparing different consumer graphs.

The validation data set cannot be used when constructing the graph itself, since doing so introduces bias into precision and recall.

Learning similarity metric

Another feature of the graph that can be adjusted as more data is introduced is the underlying similarity measure. Typically, the metric lasts for a longer period of time, e.g., 5-10 iterations of the graph, and the metric is not re-evaluated at the same frequency as the accuracy rate.

In case the distance function is not fixed, the parameters of a specific distance function may be learned, or the best distance function may be selected from a family of such functions. To learn the distance function or its parameters, the values of precision and recall are compared to the validation set.

Assume that one goal is to predict similar audience populations for high-incommers based on known attributes of the high-incommer's seed set. For different distance functions or different parameters in a particular distance function, the similarity of the seed node to all other nodes in the graph is calculated. The distance function uses the attributes of the nodes (e.g., online and television ratings) to compute the similarity.

For example, if the distance function is a radial basis function with a parameter s, where:

f(N_i,N_j)＝exp(||A_i–A_j||^2/s^2)，

then, for different values of s, the pair-wise distances of the seed node to all other nodes are calculated using the same distance threshold T to generate a set of similar nodes. For different s values (parameters to be learned), different sets of similar nodes are calculated and generated, and are represented by N _ S(s).

For the set N _ s(s), probabilities of True Positive (TP), False Positive (FP) and False Negative (FN) may be computed. True is all nodes in N _ s(s) that belong to the target set in the verification set. In this example, all nodes are also high-revenues (in the real data set). False positives are all nodes in N _ s(s) that do not belong to the target set (not high-income). False negatives are all nodes belonging to the verification set (being high-revenuers), but not to N _ s(s).

There are different tradeoffs between precision and recall depending on the application. In the case of targeting the audience of the ad, it is desirable to obtain a higher recall rate because the cost of exposure (ad) is low, whereas the cost of losing members in the targeted audience is high.

In the examples herein, the objective is to select a value of s that is high in both precision and recall from the possible values of s for other types of distance functions. For other types of distance functions, there may be other parameters that maximize precision and recall.

The accuracy of the approximation model is limited only to target audience segments. For example, television viewing and online behavior data sets may be used to predict from a high-income-people's seed set whether an approximate audience segment also includes high-income people. The predictions may be validated using a set of true revenue levels, where a set of true revenue levels is a set of nodes used to derive the predictions. This yields the accuracy of the prediction. However, for a new target portion, it makes no sense to predict the accuracy of a portion, such as whether those same users are also luxury car drivers.

Calculating a repeat arrival rate

The consumer graph connects a node (consumer) to all devices he or she uses. Thus, the graph allows for a total exposure of the individual to repeated advertisements. For example, if user abc123 had seen a particular advertisement on each of his televisions, desktops, and mobile devices, the exposure for the total repeat would be counted as 1. This allows the following metric calculations to be used to make the direct measurements.

The repeat exposure audience refers to the number of users belonging to the target audience group in the consumer graph who have been exposed to the advertisement after the repeat exposure. Then, the direct repeat arrival rate is:

repeat arrival rate-repeat viewer/total number of viewers

For spot measurement, this allows the number of spot audiences who repeatedly watch the advertisement, i.e., the number of spot users who belong to the group of target audiences who have been exposed to the advertisement after repeated exposure, to be calculated. Then, the arrival rate of the samples is:

repeat sample arrival rate (repeat sample audience/total sample audience)

For the modeled measurement data, the ID of the users from which the data was collected in the consumer graph is unknown. Therefore, the arrival rate data cannot be repeated at a one-to-one level.

If the advertiser wants to set an upper frequency limit for the consumer (e.g., if the advertiser does not want the same advertisement to be displayed to the same user more than twice), then the calculation of the repeat arrival rate is useful in managing the goals. The repeat arrival rate also provides a convenient metric for optimizing the effectiveness of an advertising campaign: for example, by calculating the repeat arrival rate over time, as the advertising campaign adjusts, improvements may continue by changing parameters of the campaign (e.g., the time and channel of the consumer population or the television content airing).

Calculating an increased arrival rate

On a certain day t, let the (direct or sampled) repeat arrival rate be x. The increased arrival rate is an additional repeat arrival rate after the activity is performed. In a cross-screen application, this is a useful parameter if an advertiser wants to be able to evaluate whether they are able to extend to 30% of reach through television and 35% of reach through mobile platforms. However, it should be noted that when measuring e.g. television data directly, the sample portion obtained for the smart tv is only a subset of the total data, since the number of smart tvs is currently relatively small in the population.

For example, where modeling measurement data is from an advertising company and the nature of the sample must be inferred, the user IDs in the consumer graph from which the data was collected are not known. Therefore, it is not possible to know whether the same user has seen the advertisement in the past. Thus, since the device cannot be associated with a particular user, increased duplicate arrival rates cannot be calculated for the modeled data. As described above, the increased repetition rate from the sampling measurements can be calculated without repetition, so the method herein is better than the layout-based method.

User equipment habit modeling

The problem of addressing the different data tracking requirements includes processing data inputs that may have ordering tags reflecting the nature of each type of device or medium. In one embodiment, the techniques herein address this issue by assigning data based on the nature of each device. For example, the first batch of consumer data is limited to data mapped from a particular device used by a particular consumer, and the sources are a plurality of third party APIs. The device data will update the consumer on each particular device as to when, where, how long, and how used. This data is then integrated and processed with other consumer data, respectively. All device usage data is integrated together so that the access point and behavior of the user over time is more accurately known.

In another example, the batch of consumer data is based on consumption data. In the data stored by each device, there is potentially included another plurality of third party data about the media actually used. Such data may be obtained from content providers, original equipment manufacturers, publishers, other data integrators, and measurement providers (e.g., Nielsen). These data provide information about what content the consumer watched. By knowing what the consumer is watching, it is possible to know the tastes and preferences of the consumer, what television programs they are watching, and when, where, and on what equipment they are watching. There are a number of ways to determine which person in a family is viewing what content (e.g., knowing which member in the family has logged into their Netflix account).

With such a structure, the system is able to integrate and process data within and across different categories. One example is to compare the complete user data set for a given consumer with the complete data sets for other members of the market segment. Each complete set of user data is cross-compared with each other set of user data. The system then matches similar behavior and can determine nuances that may affect the effectiveness of the advertisement. Such a determination also enables the predictive algorithm to be adjusted on a consumer-by-consumer basis. Thus, as described further herein, consumer data may be used for revenue optimization because the audience may be matched to television activity.

Setting up advertising campaigns

The advertiser selects various campaign parameters and campaign goals. The advertiser may select certain parameters, such as demographics, and then rate the overall population percentage that meets the campaign criteria. For example, the advertiser may be for women in california between the ages of 20 and 30, and require a 20% proportion of this population. Another criterion may be the frequency with which advertisements cover a particular population, such as "two exposures to each age group". Criteria may be narrowed to identify users who are interested in a particular product on the market, e.g., for recent searches

Women of shoes.

Taken together, the advertiser-specified criteria may be ranked and weighted by importance. For example, a woman in san Francisco may be weighted more heavily than a woman in the saxagate holder; in this case, the system may allocate and budget more exposure for a particular population in the population with higher weight. In addition, viewers can be specifically categorized according to their product purchase patterns and media consumption. Assumptions can be incorporated into the campaign parameters based on purchase history, such that individuals who purchased a luxury handbag are more inclined to react to the luxury handbag advertisement.

Next, the system can run various campaign parameters and goals against the consumer data received from the first party database and through the database of the third party API.

The underlying consumer data input is integrated from multiple sources as well as third party data sources, including in-house development data sets generated by machine learning processes. For example, a third party API representing consumer and audience data may be combined with data from a series of processes that internally process the data stream, e.g., track the behavior of the audience and may predict the audience's consumption results. Purchase recommendations may be provided and comparisons may be derived based on the relevant real-time metric application. The system may further facilitate performance of transaction bids and purchases.

In a preferred embodiment, bidding and purchasing of advertising inventory is accomplished at a faster rate than prior approaches due to the integration of data from multiple different sources in one system. This increase in speed also greatly increases the accuracy of data models that predict media consumption and consumer behavior. For example, the system can contain a series of data relating to other buyers in the trading platform to optimize purchases based on inventory allocation considerations and the like.

In a preferred embodiment, the system includes APIs from third parties that track relevant metrics such as consumer behavior and consumer demographics (e.g., age, race, location, and gender). The relevant metrics are analyzed based on the buyer's activity requirements, such as budget, desired audience and number of exposures.

To analyze existing ad inventory, the system obtains real-time inventory data through the publisher and content provider APIs. Data about inventory may be aggregated across media so that inventory available for digital, mobile, television, and OTT may be consolidated, allowing advertisers to allocate their budgets across a variety of media, device categories, and content channels. Alternatively, if the content is not uniform with the desired television slot, this integration allows the MVPD to distribute advertising content to platforms other than television.

Once the advertiser has specified campaign parameters, and the system has determined appropriate inventory, the system allows the advertiser to select strategies for optimizing exposure allocation. There are a variety of advertising strategies available to advertisers.

Example factors for advertising policy are as follows:

pace-the rate at which advertisers place advertisements.

Average pace: the activity-based budget and length are evenly distributed.

Accelerated purchase: purchase based on performance (e.g., the system detects the entire population of car video viewers and automatically assigns based on findings)

Competition pace: if an advertiser's competitors purchase a particular one of the slots in large numbers, the advertiser may choose to either compete with them or allocate from those slots that remain (this applies to any device media).

Specific time limit strategy: advertisements are purchased according to the time of day and day of the week. End-user advertisers may purchase inventory for an entire day (e.g., play every six hours) or define a specified time of day.

Inventory policy: the purchase of advertisements is based on maximum advertising expenditure or on meeting specific campaign parameters and goals.

Pricing strategy: the emphasis of the purchase is to stay within a certain budget. Budgets may allocate money based on inventory, media, and/or time limits.

Media policy: the system detects which media best meets the campaign goal.

Segmented media planning is the practice of deploying many media policies to different inventory types. Existing strategies incorporate many consumers into a high-level consumer classification.

In one example of this process, advertisers will set their media policy to target a population of middle-aged women. The system may then divide the population of middle-aged women into a plurality of sub-portions, such as women with children and women without children. Next, the system may match an advertising strategy (e.g., a balanced Pacing strategy) to an individual, such as a particular woman with children who was searching for diapers in the past hour. The system may then distribute the advertisement to a particular device of the woman, such as her cell phone, so that the next time she opens an application (e.g., YouTube), the advertisement will be displayed.

MVPD environment

As shown in FIG. 8, the techniques described herein allow the MVPD to optimize the advertiser's campaign for the available inventory. The system is used by media owners who sell various advertising opportunities to marketers who attempt to plan and deliver digital and television advertising campaigns. The broadcaster or MVPD has a database 801 of customer data and controls the system that matches advertising content to television inventory. The system accepts as input one or more campaign descriptions 127 from advertisers, an inventory of ad slots 811, a summary across multiple broadcast channels, and a set of conflict constraints 821. Conflict constraints are conditions for ad content delivery, which may come from the advertiser itself, the regulatory agency, or the MVPD itself. These conflicting constraints are "hard" constraints because the final result of matching the advertising content to a particular television slot cannot violate these constraints.

Using the methods described herein, the MVPD matches 831 the activity with the slots in inventory, subject to conflict constraints. This results in a number of possible gaps that are appropriate to match a given activity.

These slots may then undergo interactive and/or iterative steps of scene modeling (either performed automatically or processed by an operator) that utilize customer viewing information and demographic data to derive certain parameters θ. The matching process is subject to "soft" conflict constraints, e.g., by constraint satisfaction engine 851.

At the end of the process, one or more slots 861 are obtained, which slots 861 are considered the best choice for the activity. At this stage, these active slots comprise the best one, but a tradeoff may be necessary.

Satisfaction of constraint conditions

The MVPD environment may incorporate user-indicated and weighted activity constraints (e.g., hard constraints such as government regulations, and soft constraints related to business objectives) into the matching process of inventory and activity. The system derives revenue predictions and campaign success rates by incorporating priority weights provided by the users. Weight switching and control may be performed through a software interface, allowing the user to test the impact of higher or lower weights according to a list of possible constraints. The system also derives optimal weights for soft constraints that maximize MVPD revenue for ad sales and ad impressions for the available ad inventory on various tv programs hosted by the media. The advertisement putting suggestion obtained by the system effectively avoids scheduling conflict by integrating and calculating various advertisement layout constraints.

Hard constraints refer to legal (e.g., federal and state regulations) or other mandatory limitations to place an advertisement in a particular slot or a particular media. The mandatory limit may be a limit imposed by the advertiser: for example, a soft drink manufacturer may require that its advertisements not be adjacent to advertisements of a particular competing soft drink manufacturer. A legal limitation may be to require that certain categories of advertisements are not played with particular program content. (e.g., advertisements for alcoholic beverages that cannot be shown in children's programs.)

A soft constraint refers to a desired (non-mandatory) limit for placing an advertisement in a particular time slot in a particular media.

Satisfaction of a constraint refers to the process of finding a solution to a set of constraints that impose the condition that a variable must satisfy. Thus, the solution is a set of values for the variables that satisfy all of the constraints.

Multiple algorithms for solving multiple constraint satisfaction problems ("CSPs") employ a combined search and inference (constraint propagation) approach, where constraint propagation is used to reduce the search space, e.g., eliminate variable pair/value pairs that do not belong to a solution. While the ad scheduling problem is usually easily modeled by CSP, this only applies to schedules that are only subject to hard constraints. When scheduling involves soft constraints (i.e., constraints that preferably should not be violated or may be violated), CSP currently does not resolve well. Especially when all solutions violate at least one soft constraint.

The techniques herein make it possible to solve advertisement scheduling problems involving soft constraints by using one or more local csps (pcsps) or weighted csps (wcsps), where algorithms are used to minimize violation of soft constraints. Often, CSPs do not easily incorporate secondary targets (i.e., a list of business targets arranged in order or preference). In the case where there are multiple targets with weighted preferences, it is generally more efficient to use Linear Programming (LP). Thus, the underlying constraint satisfaction techniques used herein may use a combination of PCSP and linear programming to integrate hard and soft constraints. The output is an ad placement plan in available ad inventory that maximizes revenue while satisfying multiple constraints. In this way, a solution can be obtained that satisfies a subset of the total constraints.

For scheduling problems with hard and soft constraints (or weights), there is currently no general analytical solution, and another way to plan for such problems is the Weighted Constraint Satisfaction Problem (WCSP). There are several ways to resolve WCSP. For example, the local random search is valid for Max-SAT (e.g., refer to Internet site www.cs.cornell.edu/selman/papers/pdf/maxsat. pdf.). Evolutionary algorithms or random population-based optimization (e.g., methods on ieeplre. ie. org/document/6900239) can also effectively address the wider variety of constraints satisfaction issues and can be used in conjunction with the methods herein. Conditional preference networks (CP-nets) are another technique to address a mix of hard and soft constraints. WCSP yields a more efficient solution than LP in several applications, for example (see www.inra.fr/mia/T/destination/Akplogan 13. pdf).

As part of a potential revenue optimization solution, the sandbox environment (or equivalent environment isolated or isolated from external inputs) performs the following tasks, as shown in fig. 8:

(a) inputting information relating to details of the advertisement campaign and details of the advertisement copy, including any specified broadcast restriction codes;

(b) inputting data variables and limits of advertisement inventory according to a profit optimization model;

(c) obtaining an output to display the hard constrained campaign and advertisement copy; and

(d) user-selected weights are integrated into the completion, goals, and preferences of various activities through a scenario modeler.

The above-described process allows an advertising inventory seller to model the impact of different constraints' prioritization on revenue returns. The user may switch higher or lower weights across multiple named constraints.

In one example, a potential solution must place a primary airline advertisement in a commercial. To do this, the decision engine does not calculate the value of the ad placement based solely on the expected return and the fees earned per exposure. Instead, the advertisements are processed using a linear combination of advertisement attributes and priority weights. The system calculates all selections on the ranking and delivers the airline's advertisements according to the user-indicated weights. This is an improvement over the prior art methods, since the prior art methods can only calculate the benefit based on the cost incurred per exposure.

In a second example, the potential solution has a hard constraint that prohibits placing beauty product advertisements in commercials of a particular situation comedy. As described by the announcer of the program, placing this advertisement would violate the program restrictions of the situation comedy. To achieve both permissible placement and revenue optimization placement of cosmetic product advertisements to avoid violations of hard constraints, the system performs a two-stage analysis. In the first stage, the engine filters out all commercial samples that violate a univariate constraint-one of which is the program limit. In the second stage, the program generates suggestive placement results for the beauty product advertisement through the front end.

In a third example, advertisements from car manufacturers are placed in commercials. However, the content provider has a business policy that it does not play competing companies' advertisements in a continuous advertising slot. This represents a soft constraint for the content provider. The system checks constraints provided by the content provider. Content provider-user sets the weight of the constraint, for example, by using a toggle key on the user interface, which is a high-to-low slider bar set by the user. If the weight is determined to be highly preferred, the system will not create any situation where competitor ads are placed next to the car manufacturer ads. In addition, the system can also integrate hard constraints at the same time. In this example, the user provides a hard constraint rule that prohibits alcohol from being played with the car advertisement. The hard constraints are processed simultaneously by the system to limit advertising for promoting alcoholic beverages. The backend system saves the limits of all subsequent incoming ad requests to memory to ensure that the auto manufacturer and the alcoholic beverage ad requests are denied placement in the slots next to the auto manufacturer's ad.

In all embodiments of the method, the constraint satisfaction engine may solve a plurality of constraint satisfaction problems. The solution solves the binary hard constraint condition in addition to the non-binary soft constraint condition. In a third preferred embodiment involving automotive manufacturer advertising constraints, the engine's algorithm alters the placement of unauthorized advertisements, for example, placing Guinness and Ford advertisements in a gap outside of the two automotive manufacturer advertising plans. The engine can address arbitrary hard and soft constraints.

In a fourth example, the engine considers which advertisements are played during the playout of a particular children's television program. When determining which advertisement to place in a particular commercial slot, the system first filters out all advertisements according to the following implicit constraints, and treats each constraint as a hard constraint:

1. any advertisement that has reached its upper frequency limit: for example, car advertisements that have recently completed the entire purchased play amount are removed.

2. Any advertisement prohibited in the restricted placement specification for soft drink advertisements played in children's programs (e.g., advertisements promulgated by the british regulatory agency, Clearcast).

3. Any advertisement prohibited in the limited placement specification of alcoholic beverage advertisements is played in the children's program.

The system applies the above filters programmatically and the system proposes recommendations for the highest revenue ad impressions available.

In a fifth example, the system can interpret a soft constraint, which is an advertising plan goal related to the relative ordering of the customer's advertisements. This occurs after the hard constraint filter is applied and the tables are no longer authorized to deliver disallowed combinations. In this scenario, the system optimizes ad placement according to business policies related to advertiser-client management. For example, the system separates competitor's ads as much as possible without violating any hard constraints. If a user-content provider prefers this advertiser over another advertiser, the soft constraint settings may be adjusted through the front-end user interface to indicate the preference. The system interprets the preferences and places the advertisements of the more biased customers at the program break point where the objectives of the customer's campaign are best achieved. Through the front end, the user content provider may rank advertisers-clients and generate advertisement scheduling suggestions according to a preference model.

Optimizing virtual sales revenue for multiple advertising inventories

An advantage of the present technology is that a user may run multiple simulations that match advertising campaigns to inventory slots. Thus, the simulation enables revenue optimization, enabling the MVPD to evaluate which matching combinations will bring the greatest advertising revenue.

In one embodiment, the aggregate amount of advertising inventory may be divided, e.g., programmatically, into commoditized inventory pieces, which are then paired and targeted to the market segments most relevant to the respective inventory. Thus, digital advertising revenue is optimized in an automated manner through segmentation and demand pairing. Thus, the solution optimizes sales performance of inventory available to content providers (e.g., pay-tv operators, tv original equipment manufacturers, supplier-side platforms injecting ad exchange platforms, ad networks, tv and media companies selling ads, etc.).

In another embodiment, the autonomic programming system enables media vendors and content providers to maximize total profits by selectively locking out inventory buyers. The system includes a software interface through which sellers can access and manage an available inventory of advertisements. The system may also include an API product that can be used to provide data to existing enterprise platforms for either or both parties to the transaction. Thus, the system may be a stand-alone media management platform or may be a data system that interacts with an external platform.

While this system may be used by MVPDs for searching to optimize placement of advertisements in television inventory, the system may be more widely used for various cross-screen settings. To do so, it enters a list of available digital advertising inventory on a variety of devices and media. For example, in one embodiment, the system enters a number (e.g., 1000) representing the amount of exposure on a digital mobile application; one number representing exposure on the desktop site (e.g., 1000), one number representing television commercials in prime time programs (e.g., 2), and another number representing commercials in the broadcast content (e.g., 10). The system aggregates all possible side-of-sale inventory into a common relational database. The data may be structured and tagged according to its associated attributes, such as media, inventory schedules, and media broadcasts. Structuring and marking can be done automatically according to certain rules and the ability to identify content.

The system may be the same system that aggregates data related to current mainstream market conditions, such as current clearing prices, winning prices, and partner quantities in demand (buyers of inventory). More market data can be integrated into the database. These data may include historical characteristics for each market to determine the appropriate pricing models.

Through separate demander data processing, the system collects and aggregates demander (inventory buyer) data. The demander data may be generated by the demander partners through their APIs. The demander data lists all current and active requests for available advertising inventory. The system then optimizes the demander data, breaking the inventory into smaller marketable data chunks. These data blocks are sorted and rated to show an estimate of how much revenue each block will bring to the supplier's seller. The optimization process may also map the target portions from the demand parties to the available advertising inventory in the supplier data set. In one example, the demand side target section shows that the inventory buyer is interested in purchasing 1000 exposures to a 29 year old man. The system identifies the highest value for each exposure inventory from the supplier data set and automatically pushes the demand data to the supplier user.

Once paired, the system provides suggestions to the supplier user through the interface. These supplier users include content providers, original equipment manufacturers, supplier platforms such as ad exchange platforms, ad networks, television, and media companies selling ads (e.g., Fox News, CNN, SKY TV). Alternatively, the system may allocate various inventory blocks sold in various inventory markets through the API. The data of the API is optimized using the above-mentioned method and is automatically updated when new inventory becomes available. The data output of the solution is used to maximize the total revenue and provide the best match to the target part.

With the knowledge and identification of new inventory markets, the optimization tool can take into account relevant characteristics of these markets in the solution optimization process. With the advent of new media channels, such as the creation of new media consumption devices, the system will be optimized around information about consumer usage of these devices. With the addition of new demand partners, the system will recalculate the distribution of inventory sales based on the total demand of the market.

Unlike existing systems that do not autonomously interpret individual nuances in the demand market, nor automatically subdivide the demand market into individual analysis blocks, current systems benefit from automatic tagging, continuous analysis, and reorganization of available data input. The system handles a large number of possible earning opportunities (preferably including exhausting all of them) and proposes an estimate of which are the most economically valuable.

Delivering and optimizing cross-screen advertising content

The technology described herein also allows advertisers to deliver advertising content to consumers through multiple media channels, including television and online media. This is especially important in situations where the MVPD cannot find the best placement of advertising content in an existing television slot, but has access to inventory on other known devices used by the relevant consumer group.

Advertisers can target consumers in two environments. In the following steps of 1: in environment 1, the DSP need only use the actual portion and/or a simulated version of the actual portion to make real-time decisions to deliver the advertisement if the consumer matches the target parameters. In the exponential analysis, when 1: 1 targeting and no dynamic ad insertion or real-time decision making is possible, the system will look at the concentration of viewers who are expected to access the slot (e.g., television program or VOD program) and then target the slot with the highest concentration of targeted consumers.

In a preferred embodiment, the advertiser controls the distribution of advertising content in that the advertiser accesses the system through a unified interface that displays information about the inventory, manages bids for the inventory, and provides a list of potential advertising targets that are consistent with the campaign description and advertiser budget. The system then communicatively connects with the provider on the supplier side to ensure that the desired slots are purchased, typically through a bidding process, and that the advertising content is placed or scheduled for placement.

In one embodiment, the technology provides for ad placement through two or more media channels for an ad campaign, rather than placing a single ad to multiple consumers at different times (e.g., on television only). Thus, the system allows for the delivery of advertising content to a given consumer or group of consumers on multiple devices. For example, a consumer may view a portion of the activity on a television or may view the activity in a desktop browser session on their laptop or OTT device. In this case, television inventory may be purchased through various television consumer devices including, but not limited to, linear, time-shifted, traditional, and programmed televisions, according to the bidding methods described herein or by those familiar with the art. In some cases, advertisers wish to limit the number of exposures a given consumer receives; in other cases, advertisers may wish to expand advertisements from one media to another based on metrics calculated across multiple media channels. The method allows advertisers to target portions of the population more accurately than before, and enables fine-grained campaign refinement based on performance indicators of multiple channels.

This technology has two aspects that enable advertisers to successfully manage and refine advertising campaigns: the system can track which devices a given consumer can access and on which devices the user has been exposed to the advertising campaign; the system may also identify those consumers who are most likely to be interested in the active content. Thus, the accuracy of targeting can be achieved through predictions based on a mapping from aggregated cross-screen ratings data to consumer behavior.

The analysis portion of the system is capable of accepting unlimited data input regarding consumer behavior across a variety of media. The system uses these data to optimize consumer classifications. The second part of the output is to improve the measurement and prediction of future consumer behavior based on the data of cross-screen behavior.

Analysis of the cross-screen data can determine where and when a consumer views an advertisement or a particular version of an advertisement, allowing advertisers to schedule the playing of advertising campaigns on multiple platforms. The advertiser may then schedule the subsequent placement, timing, and manner of the advertisement. They may control retargeting (whether they show the same advertisement more than once) or may choose to play a multi-chapter advertisement story.

One method for managing the delivery of advertising content to consumers across more than two display devices is illustrated in FIG. 7. According to the methods described elsewhere herein, the consumer graph is a graph 710 of consumers that has been built or is being built and modified, and a pool 730 of consumers defined based on a graph of consumer attributes, where the graph of consumers contains information about the devices used by each consumer and the demographic data of each consumer, where the pool of consumers contains consumers that have at least a threshold degree of similarity to the target audience member.

The system receives a list of advertising inventory 712 from one or more media channels or content providers, where the inventory list includes one or more television slots and online slots.

The system receives price points 702 of one or more ad descriptions 705 from an advertiser, wherein each campaign description 705 includes a schedule for placement of a plurality of advertising content items on two or more devices accessed by a consumer and a target audience 720, wherein the target audience is defined by one or more demographic factors from: age range, gender and location. Price point 702 represents the advertiser's budget in an advertising campaign. The budget may be distributed across multiple slots and across multiple media channels according to active inventory and goals. The target may include a desired target audience, as well as a desired number of exposures.

Based on the pool of consumers, the campaign description, and the available inventory, the system may determine one or more advertising targets, where each of the one or more advertising targets includes two or more slots that coincide with a given price point 702 associated with the campaign description 705. The advertising content described by the one or more advertisements may then be assigned to the one or more advertising targets based on the inventory.

The above steps may be performed sequentially other than the above order, or iteratively, sequentially, or partially simultaneously. Thus, the system may receive the campaign description and price points at the same time as, or before or after, the advertisement inventory is received. In addition, the consumer graph may be continually updated.

For a given consumer, a number of devices 740 that the consumer accesses are identified. This may be accomplished by constructing a device diagram as discussed herein. Those consumers associated with multiple devices may be targeted for advertising campaigns.

The various categories of data (inventory, ad campaigns, etc.) may be input to the system through various Application Program Interfaces (APIs), the development of which is within the capabilities of those skilled in the art.

Then 770, for each slot in the advertising target, the system bids on that slot in accordance with the price point; for two slots in the bid, the system then instructs the first content provider to place the first item of advertising content in the first slot to the consumer pool on the first device, while for the second slot, the second content provider may be instructed to place the second item of advertising content in the second slot to the second device. Preferably, at least one of the first device and the second device is a television.

It should be appreciated that once the tv and online inventory slots are determined to be consistent with an advertising campaign, the steps of instructing and placing are optional to a given entity performing the method.

The methods described herein may also optimize advertising campaigns on multiple devices accessible to consumers. These methods are based on the delivery method described above and in fig. 7. Once it is determined that the consumer is a member of the target audience, and it has been determined that the consumer has access to the first and second devices, the advertiser wishes to purchase a slot for the first and second items of advertising content on the first and second devices in a manner that improves the previous advertising campaign and conforms to the advertising budget and target audience.

In this case, the system may receive feedback regarding the consumer's reactions to the first and second items of advertising content, and based on this information and similar information from other consumers, may use this feedback to indicate that more slots are being purchased for the first and second items of advertising content.

For example, the system may receive first data from a first tag accompanying a first item of advertising content to verify whether a particular consumer viewed the first item of advertising content on a first device, and second data from a second tag accompanying a second item of advertising content. The given content data may be a beacon, for example, communicated via a protocol such as VPAID or VAST.

In some embodiments, the data may include a confirmation of whether the consumer saw the first item of advertising content, in which case the second item of advertising content is not delivered to the consumer until the consumer saw the first item of advertising content.

In some embodiments, an advertising campaign may be optimized in a number of different ways. As described elsewhere herein, while a measure of the repeat arrival rate may be used to assess-and improve-the effectiveness of an advertising campaign, another factor is the overall cost-effectiveness of the advertising campaign. For example, given a budget, or a total dollar spent per ad, the cost per exposure may be calculated. This number can be optimized in successive iterations of the activity.

In other embodiments, the ad campaign is updated and optimized during the campaign. For example, an activity may be scheduled to be performed over a particular period of time, such as 3 days, 1 week, 2 weeks, 1 month, or 3 months. The system herein can provide feedback on the effectiveness of a campaign before the campaign is completed, and thus can provide advertisers with the ability and opportunity to adjust campaign parameters to improve the reach of the campaign. These parameters include, but are not limited to, aspects of the audience population such as age, income, location, and medium of advertisement delivery (e.g., television station) or time of day.

The systems and methods described herein may further provide a method (e.g., using a similarity model) for advertisers to predict future potential viewing habits based on historically accumulated audience data. The historical data may include data obtained during the course of the activity.

Computing implementation

Computer functions for processing ad campaign data, ad inventory, consumer and device maps (e.g., using bit string representations), and optimizing revenue and working under hard and soft constraints may be developed and implemented by programmers or teams of programmers in the art. These functions may be implemented in a variety of programming languages, including hybrid implementations in some cases. For example, these functions and script functions may be programmed in a functional programming language, such as: scala, golang, and R. Other programming languages may be used to implement certain portions, such as Prolog, Pascal, C + +, Java, Python, visual basic, Perl,. Net languages (e.g., C #), and other equivalent languages not listed. The capabilities of the present technology are not limited to or dependent upon the underlying programming language used for the implementation or control of access to the basic functionality. Alternatively, this functionality may be implemented by higher level functions, such as toolkits that rely on previously developed functions to manipulate mathematical expressions (e.g., bit strings and sparse vectors).

The techniques herein may be developed to operate with any of the well-known computer operating systems currently in use, as well as other operating systems not listed herein. These operating systems include, but are not limited to: windows (including Windows XP, Windows95, Windows2000, Windows Vista, Windows 7 and Windows 8, Windows Mobile and Windows 10, and Microsoft corporation's intermediate updates, etc.); apple iOS (including iOS3, iOS4, iOS5, iOS6, iOS7, iOS8, iOS9, etc., and is constantly updated); apple's MAC operating system, such as OS9, OS10.x (including variations of "Leopard", "Snow Leopard", "Mountain Lion", and "Lion"); an Android operating system; UNIX operating systems (e.g., berkeley standard version); and the Linux operating system (e.g., available from numerous publishers or "open source" software for free).

In this regard, a given implementation relies on other software components already implemented, such as functions for operating sparse vectors and functions for computing similarity measures for vectors, which it may be assumed that a programmer of skill in the art may implement.

Further, it should be understood that executable instructions which cause a suitably programmed computer to perform the methods described herein may be stored and delivered in any suitable computer-readable format. This may include, but is not limited to, portable readable drives such as large capacity "hard drives," or "memory on board" (e.g., usb port connected to computer), internal drives of the computer, and compact disc read only memory or compact disc. It is further understood that while the executable instructions may be stored on a portable computer readable medium and delivered to the purchaser or user in such a tangible form, the executable instructions may also be downloaded to the user's computer from a remote location, such as through an internet connection that may rely to some extent on wireless technology (e.g., wifi). This aspect of the technology does not imply that the executable instructions take the form of signals or other non-tangible forms. The executable instructions may also be executed as part of a "virtual machine" implementation.

The techniques herein are not limited to a particular web browser version or type; the techniques may be implemented by one or more of the following browsers: safari, internet explorer, EDGE, Firefox, Chrome, or Opera, and any other version.

Computing device

An exemplary general purpose computing device 900 suitable for implementing the methods described herein is schematically depicted in fig. 9. Such computer devices may be located within the control range of the MVPD, for example linked to an intranet within the MVPD corporate environment.

The computer system 900 includes at least one data processing unit (CPU)922, memory 938 (typically including high-speed random access memory and non-volatile memory, such as one or more disk drives), a user interface 924, a plurality of disks 934, and at least one other communication interface connection 936 for communicating with other computers and other networks of devices over networks, including the internet, such as by high-speed network cables or wireless connections. There may be a firewall 952 between the computer and the internet. At least the CPU922, memory 938, user interface 924, disks 934 and network interface 936 are in communication with each other via at least one communication bus 933.

CPU922 optionally includes a vector processor optimized for manipulating large data vectors.

The memory 938 stores programs and data, typically including some or all of the following: an operating system 940 for providing basic system services, one or more application programs (e.g., parser program 950 and compiler, not shown in FIG. 9), a file system 942, one or more databases 944 that store advertising inventory 946, campaign descriptions 948 and other information, and an optional floating point arithmetic processor (necessary for performing high-level mathematical operations). The method of the present invention may also utilize functionality contained in one or more dynamically linked libraries, which are not shown in FIG. 9 but are stored in memory 938 or disk 934.

Databases and other programs stored in memory 938 as shown in fig. 9 may optionally be stored on disk 934, where the amount of data in the database is too large to be efficiently stored in memory 938. Rather, the database can alternatively be stored in part on one or more remote computers that communicate with computer system 900 through network interface 936.

The memory 938 is encoded with instructions for receiving input from one or more advertisers, and for calculating a similarity score between consumers. The instructions also include programmed instructions for performing one or more of parsing, calculating metrics, and various statistical analyses. In some embodiments, the sparse vector itself is not computed on computer 900, but rather the computation is performed on a different computer and communicated to computer 900, for example, through network interface 936.

In particular, various implementations of the techniques herein can be performed on computing devices of varying complexity, including (but not limited to) workstations, PCs, laptops, notebooks, tablets, netbooks, and other mobile computing devices, including cell phones, mobile phones, wearable devices, and electronic organizers. The computing device may have a suitably configured processor, including (but not limited to) graphics processors, vector processors, and math co-processors, for running software that performs the methods herein. In addition, some computing functionality is typically distributed over multiple computers, such that, for example, one computer accepts input and instructions, a second or other computer accepts the instructions over a network connection, performs processing at a remote location, and optionally transmits results or output back to the first computer.

Control of the computing device may be provided through a user interface 924, which user interface 924 may include a display, mouse 926, keyboard 930, and/or other items not shown in fig. 9, such as a track pad, track ball, touch screen, stylus, voice recognition, gesture recognition techniques, or other input such as based on user eye activity, or a sub-combination or combination of any of the above. Further, this implementation allows purchasers of advertising inventory to remotely access computer 900 over a network connection and view inventory through an interface having similar attributes as interface 924.

In one embodiment, the computing device may be configured to restrict user access, for example, by scanning a QR code, gesture recognition, biometric data entry, or password entry.

When the technology is reduced to one embodiment (e.g., one or more software modules, functions, or subroutines), the manner in which the technology operates may be a batch mode-such as batch processing on a stored database of inventory and customer data, or interaction with a user entering specific instructions for a single advertising campaign.

The results of matching the advertising inventory to the advertising campaign criteria created by the techniques herein may be displayed in a tangible form, such as a screen of one or more computer displays (e.g., a display screen), laptop displays or tablets, laptops, netbooks, or mobile phones. The results may also be printed in sheet form, stored as an electronic file on a computer readable medium, or transmitted or shared between computers, or projected onto a meeting hall's screen (e.g., during a presentation).

A tool kit: the techniques herein can be implemented in a manner that lets users (e.g., purchasers of advertising inventory) access and control basic functions that provide a critical part of advertising campaign management. Certain default settings may be built into the computer implementation, but the user may select as many functions for allocating inventory as possible, allowing the user to remove certain features or adjust the weight of those features from their consideration as desired.

The toolkit may be operated by a scripting tool, or by a graphical user interface providing touch screen selections and/or pull-down menus, which can cater to the complexity of the user. The manner in which the user accesses the underlying tools is not a limitation on the novelty, creativity, or utility of the technology.

Thus, the methods herein may be implemented on one or more computing devices having a processor configured to perform the methods and encoded as executable instructions in a computer-readable medium.

For example, the techniques herein include a computer-readable medium encoded with instructions for performing a method for distributing video advertising content to consumers on a television, the instructions comprising: instructions for receiving, from an advertiser, a price point and one or more campaign descriptions, wherein each campaign description comprises a schedule for delivering advertising content to one or more televisions visited by a consumer and a target audience, wherein the target audience is defined by one or more demographic factors; instructions for determining one or more hard constraints associated with the one or more activity descriptions; instructions that define a consumer pool based on a consumer profile, wherein the consumer profile contains information about two or more televisions and mobile devices used by each consumer, demographic and online behavior data for each consumer, and similarities between pairs of consumers, and wherein the consumer pool includes consumers having at least a threshold degree of similarity to members of the target audience; instructions to receive an inventory list from one or more content providers, wherein the inventory list includes one or more slots on a television and on-line; instructions to determine one or more advertising targets, wherein each of the one or more advertising targets comprises a series of slots that are consistent with one or more of the campaign descriptions and the one or more hard constraints and have a total cost that is consistent with the price point; executing optimized instructions to distribute advertising content described by the one or more campaigns to one or more advertising targets based on one or more soft constraints, thereby producing one or more solutions; instructions to communicate a list of one or more solutions to an advertiser, wherein a solution comprises matching the campaign description to one or more slots in the television content determined to be likely to be viewed by the consumer pool; and delivering the advertising content item to the consumers in the consumer pool through a first media channel on a television.

Accordingly, the technology herein also includes a computing device having at least one processor configured to execute instructions for implementing a distribution method for delivering video advertising content to consumers on a television, the instructions comprising: instructions for receiving, from an advertiser, a price point and one or more campaign descriptions, wherein each campaign description comprises a schedule for delivering advertising content to one or more televisions visited by a consumer and a target audience, wherein the target audience is defined by one or more demographic factors; instructions for determining one or more hard constraints associated with the one or more activity descriptions; instructions that define a consumer pool based on a consumer profile, wherein the consumer profile contains information about two or more televisions and mobile devices used by each consumer, demographic and online behavior data for each consumer, and similarities between pairs of consumers, wherein the consumer pool includes consumers having at least a threshold degree of similarity to members of the target audience; instructions to receive an inventory list from one or more content providers, wherein the inventory list includes one or more slots on a television and on-line; instructions to determine one or more advertising targets, wherein each of the one or more advertising targets comprises a series of slots consistent with one or more campaign descriptors and one or more hard constraints, and has a total cost consistent with the price point; instructions for performing an optimization of the distribution of the advertising content described by the one or more campaigns to one or more advertising targets based on one or more soft constraints, thereby producing one or more solutions; instructions to communicate a list of one or more solutions to an advertiser, wherein a solution comprises matching the campaign description to one or more slots in the television content determined to be likely to be viewed by the consumer pool; and delivering the advertising content to consumers in the consumer pool through a first media channel on a television.

Cloud computing

The methods herein may be implemented to run in the "cloud". Thus, the processes performed by one or more computer processors to implement a computer-based method need not be performed by a separate computing machine or device. The processes and computations may be distributed among multiple processors in one or more data centers that are physically located at different locations from each other. Data is exchanged with the plurality of processors using a network connection, such as the internet. Preferably, a security protocol such as encryption is used to minimize the likelihood that consumer data will be corrupted. The computations performed at one or more locations remote from an entity such as an MVPD or DSP include computing and updating a customer graph and a device graph.

Examples of the invention

Example 1 user interface

An exemplary user interface is shown in FIGS. 10A-10D, which in successive figures illustrate an interface of successive steps in a revenue optimization workflow, such as may be performed in an MVPD environment.

In FIG. 10A, details of the activity are uploaded to the system through the front end.

In FIG. 10B, a user front end is provided for organizing hard constraints associated with existing ad inventory units.

In FIG. 10C, a user front end is provided for displaying advertisement units that do not violate any of the hard constraints of FIG. 10B.

In FIG. 10D, the front end displays a system scenario modeling interface with weighted implicit soft constraints and predicted revenue under selected weights.

All references cited herein are incorporated by reference in their entirety.

The above description is intended to be illustrative of various aspects of the present technology. The examples presented herein are not intended to limit the scope of the appended claims. The invention being fully described herein, it will be apparent to those of ordinary skill in the art that modifications and improvements may be made thereto without departing from the spirit or scope of the appended claims.

Claims (10)

1. A method, comprising:

receiving first ratings data associated with a first consumer view of an advertisement of an advertising campaign;

generating a first consumer classification based on the first ratings data;

calculating a predictive model of consumer behavior according to the first consumer classification;

receiving second ratings data associated with a second consumer view of an advertisement of the advertising campaign;

generating a second consumer classification based on the second ratings data;

updating the predictive model according to the second consumer classification;

calculating the accuracy of the prediction model;

calculating promotion data for the advertising campaign;

comparing the lifting data to a predictive model;

updating the prediction model based on the compared lift data and the calculated accuracy of the prediction model; and

sending results of at least one of the first consumer classification, the second consumer classification, the predictive model, or the promotion data to a user interface.

2. The method of claim 1, wherein at least the first consumer classification is a multi-dimensional consumer classification, and wherein at least one dimension of the first consumer classification is associated with a device viewing an advertisement.

3. The method of claim 2, wherein the multi-dimensional consumer classification includes link data from each dimension of the first consumer classification, wherein each dimension represents a different device used by the first consumer to view the advertisement.

4. The method of claim 2, wherein the first consumer's device comprises at least one of a television, a personal computer, or a mobile device.

5. The method of claim 1, wherein the first consumer and the second consumer are a subset of a total audience of the advertising campaign.

6. The method of claim 1, wherein the predictive model is updated by calculating a difference between a predictive model of consumer behavior and actual consumer behavior, using a feedback loop including inputs for at least the first, second, or third ratings data, and generating an output for at least an updated predictive model of consumer behavior.

7. The method of claim 6, wherein the predictive model is updated in real-time.

8. The method of claim 1, further comprising de-duplicating the first viewer rating data of the first consumer in response to the first consumer viewing the first portion of the advertisement on the first device and viewing the second portion of the advertisement on the second device.

9. The method of claim 8, wherein the first portion of the advertisement comprises: an advertisement segment comprising less than the full length of an advertisement run time or a subset of advertisements of a plurality of advertisements intended for viewing by the first consumer in the advertisement campaign.

10. A system, comprising:

at least one computer-readable medium configured to store instructions; and

at least one processor coupled to the computer-readable medium, the at least one processor configured to execute the instructions to cause the processor to perform operations, the operations comprising the steps of claims 1-9.

Images