WO2017121314A1 - 信息推荐方法及装置 - Google Patents

信息推荐方法及装置 Download PDF

Info

Publication number
WO2017121314A1
WO2017121314A1 PCT/CN2017/070764 CN2017070764W WO2017121314A1 WO 2017121314 A1 WO2017121314 A1 WO 2017121314A1 CN 2017070764 W CN2017070764 W CN 2017070764W WO 2017121314 A1 WO2017121314 A1 WO 2017121314A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
recommendation
recommended
click
click rate
Prior art date
Application number
PCT/CN2017/070764
Other languages
English (en)
French (fr)
Inventor
吕培立
Original Assignee
腾讯科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Priority to KR1020177034700A priority Critical patent/KR101999471B1/ko
Priority to JP2017562719A priority patent/JP6588572B2/ja
Publication of WO2017121314A1 publication Critical patent/WO2017121314A1/zh
Priority to US15/825,396 priority patent/US11301525B2/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2457—Query processing with adaptation to user needs
    • G06F16/24578—Query processing with adaptation to user needs using ranking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0242—Determining effectiveness of advertisements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/535—Tracking the activity of the user

Definitions

  • the present invention relates to the field of information recommendation technologies, and in particular, to a method and device for information recommendation.
  • the information recommendation means that among the plurality of candidate recommendation information, the recommendation information with a high click rate is selected and pushed to the user, so that the information after the push has a high degree of matching with the user, thereby improving the effectiveness of the information recommendation.
  • a typical information recommendation such as an advertisement recommendation, is a series of candidate advertisements that are placed on an advertiser, sorted by click rate, thereby pushing an advertisement with a high click rate to the user.
  • the factors affecting the click probability of the recommended information may be referred to as recommendation features.
  • the recommended features are mainly users, information (such as advertisements), information display positions (such as advertising spaces), and various types of recommended features include specific content.
  • the main purpose is to establish a click-rate estimation model including three types of recommended features: user, information, and information display.
  • the rate estimation model the click rate after the push of the recommended information is estimated.
  • the embodiment of the present invention provides a method and device for recommending information, so that Accurately select the recommended features of the click-through rate prediction model and improve the accuracy of subsequent click-through rate estimates.
  • the embodiments of the present invention provide the following technical solutions.
  • an information recommendation method comprising:
  • the click information of the recommendation information is poor
  • the target push information pushed to the user to be recommended is determined based on the click rate estimation model.
  • an information recommendation apparatus comprising:
  • a candidate factor determining module configured to determine at least one candidate recommended feature factor
  • Clicking the information amount difference determining module configured to estimate the click information amount difference of the recommendation information before and after the click rate estimation model is added to the candidate recommendation factor;
  • a feature selection module configured to determine a recommended feature factor that the click information amount difference meets the set amount difference condition, and add the determined recommended feature factor to the click rate estimation model
  • the push information determining module determines the target push information pushed to the user to be recommended based on the click rate estimation model.
  • a non-volatile storage medium for storing one or more computer programs, wherein the computer program comprises processor-executable instructions having one or more memories; The processor runs the instruction to perform the above information recommendation method.
  • the embodiment of the present invention can determine the click rate of the recommended feature factor on the recommended information by estimating the click information amount difference of the recommendation information before and after the estimated click rate estimation model is added, thereby clicking the information.
  • the recommended feature factor of the quantity difference is in accordance with the set amount difference condition, and is determined as a recommended feature factor having a large discriminating click rate of the recommended information, and determining that the recommended feature factor needs to be added to the click rate estimation model;
  • the recommended feature factor to be added to the click-rate estimation model has a large click-through rate discrimination for the recommended information, so the selected recommended feature factor has a greater influence on the click-through rate estimation, and the click-through rate is estimated.
  • the most influential recommendation feature factor is added to the click-through rate estimation model to achieve a more accurate selection of click-through rate estimates.
  • the recommended feature of the model, the recommended feature factor of the click-rate estimation model is used as the reference basis for selecting the recommended features in the subsequent click-through rate estimation, which provides the possibility to improve the accuracy of the click-through rate estimation.
  • FIG. 1 is a flowchart of a method for recommending information according to an embodiment of the present invention
  • FIG. 2 is another flowchart of an information recommendation method according to an embodiment of the present invention.
  • FIG. 3 is another flowchart of determining a target push information to be pushed to a user to be recommended in the information recommendation method according to an embodiment of the present disclosure
  • Figure 4 is a schematic diagram of advertisement push
  • FIG. 5 is a structural block diagram of an information recommendation apparatus according to an embodiment of the present invention.
  • FIG. 6 is a structural block diagram of a click information quantity difference determining module according to an embodiment of the present invention.
  • FIG. 7 is a structural block diagram of an information entropy determining unit according to an embodiment of the present invention.
  • FIG. 8 is a structural block diagram of an information entropy reduction determining unit according to an embodiment of the present invention.
  • FIG. 9 is another structural block diagram of a click information amount difference determining module according to an embodiment of the present invention.
  • FIG. 10 is a structural block diagram of a message recommendation device push information determining module according to an embodiment of the present invention.
  • FIG. 11 is a block diagram showing the hardware structure of an electronic device according to an embodiment of the present invention.
  • the method is applicable to an electronic device having data processing capability, and is preferably a recommendation server that can perform information recommendation. Referring to FIG. 1, The method can include:
  • Step S100 Determine at least one candidate recommendation feature factor.
  • the recommended feature factor is the specific content included in a certain type of recommendation feature.
  • the recommended feature of the user class may include recommended feature factors such as age, gender, region, used Internet device, historical click, and interest reflected by the purchase behavior.
  • the candidate recommendation feature factor is a recommended feature factor that can be added to the click rate estimation model to be analyzed.
  • the determined at least one candidate recommendation feature factor may belong to the same type of recommendation feature, such as a recommendation feature belonging to the user class.
  • the determined at least one candidate recommendation feature factor may also be that a part of the recommended feature factors belong to the recommended feature of the user class, and another part of the recommended feature factors belong to the recommended feature of the information display class.
  • Step S110 Before the estimated click rate estimation model is added to the candidate recommendation factor of the candidate, the amount of click information of the recommendation information is poor.
  • the principle of selecting the recommended feature factor in the embodiment of the present invention is that the recommended feature factor that has a large influence on the click rate of the recommendation information is determined to be added to the click rate estimation model.
  • gender as a recommendation feature factor under the recommendation feature of a user class can predict the effect of gender on the click rate of recommendation information, that is, when considering the recommendation feature factor of gender and not considering the recommendation feature factor of gender.
  • the difference in the click-through rate of the recommended information When the gender has a large difference in the click-through rate of the recommended information, the gender is considered to be a recommended feature factor that has a large influence on the click-through rate of the recommended information.
  • the influence of the recommended feature factor on the click rate of the recommended information can be measured by the click rate estimation model before and after the recommended feature factor is added, and the amount of click information of the recommended information is measured.
  • Step S120 Determine a recommended feature factor that the click information amount difference meets the set condition, and add the determined recommended feature factor to the click rate estimation model.
  • the recommended information of the click information difference is determined as the click rate of the recommended information.
  • the recommended feature factor is affected, so that the recommended feature factor is determined to be added to the click rate estimation model, that is, the click rate estimation model needs to consider the recommended feature factor when estimating the click rate after the recommended information is pushed.
  • the setting condition may be set, for example, the threshold value is set, and the recommended feature factor that meets the set condition of the click information difference is determined to be added to the click rate estimation model; if the click information difference is greater than the set amount
  • the recommended feature factor of the difference threshold needs to be added to the click rate prediction model.
  • each recommended feature factor may be sorted according to the difference in click information corresponding to each recommended feature factor, so that the recommended feature factor with the sequence position in the set range is determined to be added to the click rate pre- Estimate the model.
  • Step S130 Determine, according to the click rate estimation model, target push information pushed to the user to be recommended.
  • the information recommendation method provided by the embodiment of the present invention includes: determining at least one candidate recommendation feature factor; for each recommended feature factor, before and after the estimated click rate estimation model is added to the recommended feature factor, the click information of the recommendation information is poor; determining the click The recommended feature factor whose information amount difference meets the set amount difference condition is added to the click rate estimation model, thereby determining the target push information pushed to the user to be recommended based on the click rate estimation model.
  • the embodiment of the present invention can determine the click rate of the recommended feature factor on the recommended information by estimating the click information of the recommended information before and after the recommended click factor estimation model is added to the recommended feature factor, thereby determining the click information amount of the recommended feature information.
  • the recommended feature factor that meets the set difference condition is determined as a recommended feature factor that is highly distinguishable from the click rate of the recommended information, and determines that the recommended feature factor needs to be added to the click rate estimation model; as determined by the embodiment of the present invention
  • the recommended feature factor to be added to the click-rate estimation model has a large click-through rate discrimination for the recommended information. Therefore, the selected recommended feature factor has a greater impact on the click-through rate estimation, and the impact on the click-through rate prediction is affected.
  • the larger recommended feature factor is added to the click-rate estimation model, which can achieve more accurate selection of the recommended features of the click-through rate estimation model.
  • the recommended feature factor of the click-rate estimation model is used as the recommended feature for the subsequent click-through rate estimation.
  • the basis for the reference is to improve the accuracy of the click-through rate estimate.
  • the click information of the recommendation information is poor, and the recommendation information may be added before and after the recommended feature factor by the click rate estimation model, and the recommendation information is characterized by the increase of the click; and the click rate estimation Before and after the model is added to the recommended feature factor, the amount of increase in the recommended information is clicked before and after the recommended feature factor is added by the click rate prediction model, and the information entropy reduction amount of the recommendation information is expressed, that is, the information entropy reduction amount can be regarded as a click. The increase in the recommended information is clicked before and after the rate estimation model is added to the recommended feature factor.
  • FIG. 2 shows another flowchart of the information recommendation method provided by the embodiment of the present invention.
  • the method may include:
  • Step S200 Determine at least one candidate recommendation feature factor.
  • Step S210 Before the estimated click rate estimation model is added to the candidate recommendation factor, the first information entropy of the recommendation information, and the click rate estimation model are added to the candidate recommendation factor, and the second information entropy of the information is recommended.
  • the first information entropy of the recommendation information may be estimated before the click rate estimation model is added to the recommended feature factor according to the formula ⁇ f ⁇ p(f)H(y
  • f is a recommended feature factor and y is a set of target values for which the recommended information is clicked and not clicked
  • p(f) is the probability of occurrence of the candidate recommended feature factor f
  • f) - ⁇ y p(y
  • f) is The conditional probability of y and f, such as the probability that the recommendation information is clicked under f, and the probability of not being clicked.
  • an advertisement ad, a target value set y, y representing a user clicking or not clicking on the advertisement may be a set of 1 and -1, 1 may indicate that the user clicks on the advertisement, and -1 may It is indicated that the user does not click on the advertisement, and obviously can also set 1 to indicate that the user does not click on the advertisement, and -1 to indicate that the user clicks on the advertisement; in the embodiment of the present invention, the click rate prediction model can be calculated by f ⁇ p(f)H(y
  • the embodiment of the present invention may be recommended according to the formula - ⁇ f,ad ⁇ p(f,ad)H(y
  • Second information entropy of information
  • p(f, ad) is the joint probability of occurrence of f and ad, such as the probability that the recommendation information exists simultaneously with f
  • f, ad) - ⁇ y p(y
  • f,ad) is the conditional probability of y and f and ad, such as the probability of the user clicking on the ad for f, and not clicking on it The probability of ad.
  • the embodiment of the present invention may be - ⁇ f,ad ⁇ p(f,ad)H(y
  • Step S220 Estimating, according to the first information entropy and the second information entropy, an information entropy reduction amount of the recommendation information before and after the click rate estimation model is added to the recommended feature factor.
  • the recommendation information is corresponding to the increase amount of the click.
  • Step S230 The recommended feature factor that determines that the information entropy reduction amount meets the set information entropy reduction condition is added to the click rate estimation model.
  • the set information entropy reduction condition is an optional form of the set amount difference condition.
  • the recommended feature factor whose information entropy reduction amount is less than the set information entropy reduction value may be determined to be added to the click. Rate estimation model.
  • the recommended feature factors are sorted according to the information entropy reduction amount of each recommended feature factor, so that the recommended feature factor with the sequence position in the set range is determined to be added to the click rate pre- Estimate the model.
  • the online advertising system quickly and automatically finds the best match between the advertising, the advertising environment and the audience through the algorithm set.
  • the realization of this automatic optimal matching is the result of interaction of various program algorithms such as data mining, information retrieval, text analysis, emotional computing, and machine learning.
  • the system begins the mathematical simulation of the user model, uses the collaborative filtering technology to determine the audience's interest model, uses the database retrieval technology to find the best advertisement, and uses text analysis technology to determine the correlation between the advertisement and the placement environment.
  • Degree combined with the logistic regression model to predict the user click probability, to sort the ads, and finally achieve the precise delivery of the ads.
  • the characteristics of the click rate estimation model include three aspects: users, advertising spaces, and advertisements.
  • the characteristics of the user side generally include the user's age, gender, region, use of the Internet device, history clicks, and interest reflected by the purchase behavior.
  • User-side features are generally used to finely group users and learn the user group's preference for advertisements. .
  • the characteristics of the ad slot side generally include the location, size, context and other characteristics of the ad slot.
  • the click rate of different ad slots often has a natural difference. Different contexts in the ad slot often play a significant role in the click rate of different ads. .
  • the information gain of the user-side feature reflects the difference in the click-through rate of the clicks of different user groups. Different, the same reason, the information gain of the advertising space feature also reflects the influence of different advertising space characteristics on the advertising click rate.
  • the overall click rate of a certain group of people and an advertisement slot feature is not estimated, but the specific recommendation factor of the user feature and the advertisement position feature is used to estimate the click rate. To achieve accurate estimates of the clickthrough rate of each ad.
  • the estimated click rate is estimated.
  • the information entropy reduction of the advertisement is the information gain brought by the advertisement ad recommendation feature factor f. Similar processing is performed on each candidate recommendation factor, and the information entropy reduction corresponding to each recommended feature factor is obtained, so that the information entropy reduction amount conforms to the recommended feature factor of the set information entropy reduction condition, and is determined to be added to the click rate.
  • the estimation model is performed by the model.
  • the embodiment of the present invention performs a priori estimation of the click rate of the recommended information by recommending the feature factor, thereby determining the recommended feature factor with large discriminating click rate, and realizing a more accurate selection of the click rate prediction model.
  • the recommended features provide the possibility to improve the accuracy of subsequent click-through rate estimates.
  • the posterior probability of the click behavior for a recommendation information ad may be calculated as follows:
  • the embodiment of the present invention can be based on a formula Predicting the amount of information entropy reduction of the recommendation information before and after the click rate estimation model is added to the recommended feature factor;
  • IG is the information entropy reduction amount
  • f is the recommended feature factor
  • ad is the recommendation information
  • y is the target value set in which the recommendation information is clicked and not clicked
  • p(f, ad, y) is f, ad
  • f, ad) is the conditional probability of y and f and ad
  • f) is the conditional probability of y and f.
  • the information recommendation method provided by the embodiment of the invention can select the recommended features of the click rate prediction model more accurately, and provides the possibility of improving the accuracy of the subsequent click rate estimation.
  • the embodiment of the present invention may perform information recommendation, such as advertisement recommendation.
  • FIG. 3 is a flow chart showing the steps of determining the target push information to be pushed to the user to be recommended in the information recommendation method according to the embodiment of the present invention.
  • the step may include:
  • Step S300 Retrieving a click rate prediction model and at least one candidate recommendation information, wherein the click rate prediction model is added with a certain recommended feature factor.
  • the determined recommendation feature factor may be used as a reference feature for selecting a feature used in the selection of the click rate prediction model when the information is recommended; and thus, the specific recommendation information and the characteristics of the user to be recommended according to the specific recommendation information and the user to be recommended.
  • the user feature factor and the information feature factor corresponding to the reference feature are determined, and then the click rate prediction function model is used to estimate the click rate of the user feature factor and the information feature factor.
  • Step S310 Determine a user to be recommended, and obtain a user feature factor corresponding to the at least one recommended feature factor among the user features of the user to be recommended.
  • Step S320 Determine, for each candidate recommendation information, an information feature factor corresponding to the at least one recommended feature factor.
  • Step S330 Determine, for each candidate recommendation information, a click rate corresponding to the user feature factor and the corresponding information feature factor by using the click rate prediction model.
  • Step S340 Determine target recommendation information to be pushed to the user to be recommended from the recommendation information of the at least one candidate according to the click rate of the recommendation information of each candidate.
  • the recommendation information of each candidate may be sorted according to the click rate of each candidate recommendation information, so that the recommendation information in which the sequence is in the setting requirement is determined to be pushed to the user to be recommended.
  • Target recommendation information may be sorted according to the click rate of each candidate recommendation information, so that the recommendation information in which the sequence is in the setting requirement is determined to be pushed to the user to be recommended.
  • Step S350 Push the target recommendation information to the user to be recommended.
  • FIG. 4 shows a corresponding schematic diagram of the advertisement push.
  • the ad pool is used to store creatives provided by advertisers, and so on.
  • the delivery system is used to accept the request to display the page, and according to the user's information, the advertisement corresponding to the user orientation is pulled from the advertisement pool, and the click rate provided by the click prediction model is used for delivery.
  • the click rate estimation model is used to predict the probability that the user clicks on the advertisement on the page when the user has access to the advertisement page, and the user who visits the advertisement page can be regarded as the user to be recommended; the simplest
  • the click rate prediction model is a logistic regression model, which can be composed of user u, display page d, and the characteristics of advertisement a are combined into one feature vector. Let the user's click behavior y ⁇ (0,1), logistic regression can be written as:
  • the model parameter w can be solved by applying gradient random drop (SGD); for new user requests, it can be passed The probability p 0 of the user's click behavior in the request is predicted.
  • SGD gradient random drop
  • the feature selection is used to determine the recommended feature factors of the candidate based on the information of the user's click on the advertisement, the user feature, and the like, and the recommended feature factor for the click rate of the advertisement is analyzed according to the information recommendation method provided by the embodiment of the present invention. Therefore, it is added to the click-rate estimation model as a click-rate estimation model to select the recommended features when estimating the click-through rate of the advertisement.
  • Clicking on the exposure log The user's click on the ad and the exposure behavior are recorded in the form of a log. Click the content of the exposure log such as the user ID, click or the time when the exposure occurred, and so on.
  • the user portrait is used to analyze each user's specific tag by analyzing massive user basic information data and click behavior data to mark the user's interests, behaviors and other characteristics.
  • the recommendation feature of the click rate prediction model is selected more accurately, so that the accuracy of the user's click recommendation information estimated by the click rate estimation model is higher, and the recommendation information is more accurate.
  • the information recommendation device provided by the embodiment of the present invention is introduced below, and the information described below is pushed.
  • the recommended device can be referred to each other in correspondence with the information recommendation method described above.
  • FIG. 5 is a structural block diagram of an information recommendation apparatus according to an embodiment of the present invention.
  • the apparatus is applicable to an electronic device having data processing capability, and is preferably a recommendation server that can perform information recommendation.
  • the information recommendation device may include:
  • a candidate factor determining module 100 configured to determine at least one candidate recommended feature factor
  • Clicking the information amount difference determining module 110 configured to predict, before and after the click rate estimation model is added to the recommended feature factor of the candidate, the click information amount difference of the recommendation information;
  • the feature selection module 120 is configured to determine a recommended feature factor that the click information amount difference meets the set condition, and add the determined recommended feature factor to the click rate estimation model;
  • the push information determining module 130 determines the target push information pushed to the user to be recommended based on the click rate estimation model.
  • FIG. 6 shows an optional structure of the click information difference determining module 110 according to the embodiment of the present invention.
  • the click information difference determining module 110 may include:
  • the information entropy determining unit 111 is configured to: before the click rate estimation model is added to the recommended feature factor of the candidate, the first information entropy of the recommendation information, and the click rate estimation model are added to the recommended feature factor of the candidate, and then recommended Second information entropy of information;
  • the information entropy reduction determining unit 112 is configured to estimate, according to the first information entropy and the second information entropy, that the information entropy of the recommendation information is reduced before and after the click rate estimation model is added to the recommended feature factor of the candidate.
  • the information entropy reduction amount and the click rate estimation model are added before and after the candidate recommendation factor of the candidate, and the recommendation information is corresponding to the increase amount of the click.
  • FIG. 7 shows an optional structure of the information entropy determining unit 111 according to the embodiment of the present invention.
  • the information entropy determining unit 111 may include:
  • a first information entropy determining subunit 1111 configured to predict, according to the formula ⁇ f ⁇ p(f)H(y
  • f is a recommended feature factor and y is a set of target values for which the recommended information is clicked and not clicked
  • p(f) is the probability of occurrence of the recommended feature factor f
  • f) - ⁇ y p(y
  • f) is y and f Conditional Probability.
  • a second information entropy determining subunit 1112 configured to predict, after the click rate prediction model is added to the candidate recommendation factor according to the formula - ⁇ f,ad ⁇ p(f,ad)H(y
  • p(f, ad) is the joint probability of occurrence of f and ad
  • f,ad) - ⁇ y p(y
  • f, ad) is the conditional probability of y and f and ad.
  • FIG. 8 shows an optional structure of the information entropy reduction determining unit 112.
  • the information entropy reduction determining unit 112 may include:
  • IG is the information entropy reduction amount.
  • FIG. 9 shows another optional structure of the click information difference determining module 110 according to the embodiment of the present invention.
  • the click information difference determining module 110 may include:
  • a calculation unit 113 for using a formula according to Estimating the information entropy reduction amount of the recommendation information before and after the click rate estimation model is added to the candidate recommendation factor;
  • IG is the information entropy reduction amount
  • f is a recommended feature factor
  • ad is recommendation information
  • y is a target value set in which recommended information is clicked and not clicked
  • p(f,ad,y) is the joint probability of f,ad and y
  • f,ad) is the conditional probability of y and f and ad
  • f) is the conditional probability of y and f .
  • the information recommendation apparatus provided by the embodiment of the invention can select the recommended features of the click rate prediction model more accurately, and provides the possibility of improving the accuracy of the subsequent click rate estimation.
  • an information recommendation apparatus is provided, and the information recommendation apparatus described below and the information recommendation method described above correspond to each other.
  • FIG. 10 is a structural block diagram of a push information determining module 130 in an information recommendation apparatus according to an embodiment of the present invention.
  • the push information determining module 130 may include:
  • the retrieving unit 200 is configured to retrieve a click rate prediction model and at least one candidate recommendation information, wherein the click rate estimation model is added to the determined at least one recommended feature factor;
  • the user feature factor determining unit 210 is configured to determine a user to be recommended, and obtain the user to be recommended a user feature factor corresponding to the at least one recommended feature factor;
  • the information feature factor determining unit 220 is configured to determine, for the recommendation information of each candidate, an information feature factor corresponding to the at least one recommended feature factor;
  • the click rate estimation unit 230 is configured to determine a click rate corresponding to the user feature factor and the corresponding information feature factor by using the click rate prediction model for the recommendation information of each candidate;
  • the target recommendation information determining unit 240 is configured to determine target recommendation information to be pushed to the user to be recommended from the recommendation information of the at least one candidate according to a click rate of the recommended information of each candidate;
  • the pushing unit 250 is configured to push the target recommendation information to the user to be recommended.
  • FIG. 11 is a block diagram showing a hardware structure of an electronic device loaded with an information recommendation device, such as a recommendation server, referring to FIG. 11, the electronic device may include: a processor 1, a communication interface 2, a memory 3, and Communication bus 4;
  • the processor 1, the communication interface 2, and the memory 3 complete communication with each other through the communication bus 4;
  • the communication interface 2 can be an interface of the communication module, such as an interface of the GSM module;
  • a processor 1 for executing a program
  • a memory 3 for storing a program
  • the program can include program code, the program code including computer operating instructions.
  • the processor 1 may be a central processing unit CPU, or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.
  • CPU central processing unit
  • ASIC Application Specific Integrated Circuit
  • the memory 3 may include a high speed RAM memory and may also include a non-volatile memory such as at least one disk memory.
  • the program can be specifically used to:
  • the click information of the recommendation information is poor
  • the target push information pushed to the user to be recommended is determined based on the click rate estimation model.
  • the steps of a method or algorithm described in connection with the embodiments disclosed herein can be implemented directly in hardware, a software module executed by a processor, or a combination of both.
  • the software module can be placed in random access memory (RAM), memory, read only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or technical field. Any other form of storage medium known.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Strategic Management (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Operations Research (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Algebra (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种信息推荐方法及装置，其中方法包括：确定至少一个候选的推荐特征因子（S100）；预估点击率预估模型加入该候选的推荐特征因子前后，推荐信息的点击信息量差（S110）；确定点击信息量差符合设定条件的推荐特征因子，并将所述确定的推荐特征因子加入所述点击率预估模型（S120）；以及基于所述点击率预估模型确定向待推荐用户推送的目标推送信息（S130）。

Description

信息推荐方法及装置
本申请要求于2016年1月12日提交中国专利局，申请号为201610017754.7，发明名称为“一种推荐特征确定方法、信息推荐方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及信息推荐技术领域，具体涉及一种信息推荐方法及装置。
背景技术
信息推荐是指在众多候选的推荐信息中，选择点击率高的推荐信息推送给用户，以使推送后的信息与用户间具有较高的匹配程度，提升信息推荐的有效性。典型的信息推荐如广告推荐，即对广告主投放的一系列候选广告，按照点击率进行排序，从而将点击率高的广告推送给用户。
为提升信息推荐的有效性，在信息推荐的过程中，预估用户对推荐信息的点击概率尤为必要。而影响推荐信息点击概率的因素可以称为推荐特征，推荐特征主要有用户、信息(如广告)、信息展示位(如广告位)三类，各类推荐特征下又包含有具体内容。目前主要通过建立一个包含用户、信息、信息展示位三类推荐特征的点击率预估模型，进而在信息推荐的过程中，针对具体的待推荐信息及待推荐用户，选择可用的推荐特征加入点击率预估模型中，预估出该待推荐信息推送后的点击率。
可以看出，为提升点击率预估的准确性，在点击率的预估过程中，选择什么样的推荐特征加入点击率预估模型尤为重要；目前在选择加入点击率预估模型的推荐特征时，主要是基于工作人员的主观分析，并没有较为准确的推荐特征选择方案，因此需要一种信息推荐方法，以较为准确的选择加入点击率预估模型的推荐特征，提升后续点击率预估的准确性，显得尤为必要。
发明内容
有鉴于此，本发明实施例提供一种信息推荐方法及装置，从而可以较为 准确的选择加入点击率预估模型的推荐特征，并提高了后续点击率预估的准确性。
为实现上述目的，本发明实施例提供如下技术方案。
根据本发明的一个方面，提供了一种信息推荐方法，其包括：
确定至少一个候选的推荐特征因子；
预估点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的点击信息量差；
确定点击信息量差符合设定量差条件的推荐特征因子，并将所述确定的推荐特征因子加入所述点击率预估模型；以及
基于所述点击率预估模型确定向待推荐用户推送的目标推送信息。
根据本发明的另一方面，提供了一种信息推荐装置，包括：
候选因子确定模块，用于确定至少一个候选的推荐特征因子；
点击信息量差确定模块，用于预估点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的点击信息量差；
特征选择模块，用于确定点击信息量差符合设定量差条件的推荐特征因子，并将所述确定的推荐特征因子加入所述点击率预估模型；以及
推送信息确定模块，基于所述点击率预估模型确定向待推荐用户推送的目标推送信息。
根据本发明的另一方面，提供了一种非易失性存储介质，用于存储一个或多个计算机程序，其中，所述计算机程序包括具有一个或多个存储器的处理器可运行的指令；所述处理器运行该指令以执行上述信息推荐方法。
基于上述技术方案，本发明实施例可以通过预估点击率预估模型加入推荐特征因子前后，推荐信息的点击信息量差，来判断推荐特征因子对推荐信息的点击率区分性，从而将点击信息量差符合设定量差条件的推荐特征因子，确定为对推荐信息的点击率区分性大的推荐特征因子，并确定该推荐特征因子需加入点击率预估模型中；由于本发明实施例确定的需加入点击率预估模型的推荐特征因子，具有对推荐信息的较大点击率区分性，因此所选择的推荐特征因子对于点击率预估的影响较大，将该对点击率预估的影响较大的推荐特征因子加入点击率预估模型中，可实现较为准确的选择加入点击率预估 模型的推荐特征，加入点击率预估模型的推荐特征因子作为后续点击率预估时选取推荐特征的参考依据，为提升点击率预估的准确性提供了可能。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据提供的附图获得其他的附图。
图1为本发明实施例提供的信息推荐方法的流程图；
图2为本发明实施例提供的信息推荐方法的另一流程图；
图3为本发明实施例提供的信息推荐方法中确定向待推荐用户推送的目标推送信息步骤的另一流程图；
图4为广告推送的示意图；
图5为本发明实施例提供的信息推荐装置的结构框图；
图6为本发明实施例提供的点击信息量差确定模块的结构框图；
图7为本发明实施例提供的信息熵确定单元的结构框图；
图8为本发明实施例提供的信息熵减小确定单元的结构框图；
图9为本发明实施例提供的点击信息量差确定模块的另一结构框图；
图10为本发明实施例提供的信息推荐装置推送信息确定模块的结构框图；以及
图11为本发明实施例提供的电子设备的硬件结构框图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
图1为本发明实施例提供的信息推荐方法的流程图，该方法可应用于具有数据处理能力的电子设备，优选为可进行信息推荐的推荐服务器；参照图1， 该方法可以包括：
步骤S100、确定至少一个候选的推荐特征因子。
推荐特征因子为某一类推荐特征下包含的具体内容，如用户类的推荐特征下可包含年龄、性别、地域、使用的上网设备、历史点击及购买行为所反应出的兴趣等推荐特征因子。
在本发明实施例中，候选的推荐特征因子为待分析的是否可以加入点击率预估模型的推荐特征因子。
可选的，所确定的至少一个候选的推荐特征因子，可以属于同一类型的推荐特征，如同属于用户类的推荐特征。可选地，所确定的至少一个候选的推荐特征因子，也可以是一部分推荐特征因子属于用户类的推荐特征，另一部分推荐特征因子属于信息展示位类的推荐特征。
步骤S110、预估点击率预估模型加入该候选的推荐特征因子前后，推荐信息的点击信息量差。
本发明实施例选择推荐特征因子的原则为：将对推荐信息的点击率影响大的推荐特征因子，确定为需要将其加入点击率预估模型。
例如，将性别作为一种用户类的推荐特征下的推荐特征因子，可以预估性别对于推荐信息点击率的影响，即在考虑性别这一推荐特征因子时和不考虑性别这一推荐特征因子时，推荐信息点击率的差异，当性别对于推荐信息点击率的差异较大时，则认为性别是一种对推荐信息的点击率影响大的推荐特征因子。
推荐特征因子对推荐信息的点击率影响，可以通过点击率预估模型加入推荐特征因子前后，推荐信息的点击信息量差来衡量。
步骤S120、确定点击信息量差符合设定条件的推荐特征因子，并将所确定的推荐特征因子加入所述点击率预估模型。
在预估到点击率预估模型加入推荐特征因子前后，推荐信息的点击信息量差后，本发明实施例可将点击信息量差较大的推荐特征因子，认定为是对推荐信息的点击率影响大的推荐特征因子，从而将该推荐特征因子确定为需加入点击率预估模型，即点击率预估模型在预估推荐信息推送后的点击率时需考虑该推荐特征因子。
本发明实施例可设置设定条件，如设定量差阈值，将点击信息量差符合设定条件的推荐特征因子确定为加入所述点击率预估模型；如点击信息量差大于设定量差阈值的推荐特征因子需加入所述点击率预估模型。可选的，本发明实施例也可根据各推荐特征因子对应的点击信息量差，对各推荐特征因子进行排序，从而将序位处于设定范围的推荐特征因子，确定为需加入点击率预估模型。
步骤S130、基于所述点击率预估模型确定向待推荐用户推送的目标推送信息。
本发明实施例提供的信息推荐方法，包括：确定至少一个候选的推荐特征因子；对于各推荐特征因子，预估点击率预估模型加入推荐特征因子前后，推荐信息的点击信息量差；确定点击信息量差符合设定量差条件的推荐特征因子加入所述点击率预估模型，从而基于点击率预估模型确定向待推荐用户推送的目标推送信息。
可以看出，本发明实施例可以通过预估点击率预估模型加入推荐特征因子前后，推荐信息的点击信息量差，来判断推荐特征因子对推荐信息的点击率区分性，从而将点击信息量差符合设定量差条件的推荐特征因子，确定为对推荐信息的点击率区分性大的推荐特征因子，并确定该推荐特征因子需加入点击率预估模型中；由于本发明实施例确定的需加入点击率预估模型的推荐特征因子，具有对推荐信息的较大点击率区分性，因此所选择的推荐特征因子对于点击率预估的影响较大，将该对点击率预估的影响较大的推荐特征因子加入点击率预估模型中，可实现较为准确的选择加入点击率预估模型的推荐特征，加入点击率预估模型的推荐特征因子作为后续点击率预估时选取推荐特征的参考依据，为提升点击率预估的准确性提供了可能。
可选的，点击率预估模型加入推荐特征因子前后，推荐信息的点击信息量差，可以通过点击率预估模型加入推荐特征因子前后，推荐信息被点击的增加量表征；而点击率预估模型加入推荐特征因子前后，推荐信息被点击的增加量，可以通过点击率预估模型加入推荐特征因子前后，推荐信息的信息熵减小量表示，即所述信息熵减小量可以认为是点击率预估模型加入推荐特征因子前后，推荐信息被点击的增加量。
相应的，图2示出了本发明实施例提供的信息推荐方法的另一流程图，参照图2，该方法可以包括：
步骤S200、确定至少一个候选的推荐特征因子。
步骤S210、预估点击率预估模型加入该候选的推荐特征因子前，推荐信息的第一信息熵，及点击率预估模型加入该候选的推荐特征因子后，推荐信息的第二信息熵。
可选的，本发明实施例可根据公式∑f{p(f)H(y|f)}预估点击率预估模型加入推荐特征因子前，推荐信息的第一信息熵；
其中，f为推荐特征因子，y为推荐信息被点击和不被点击的目标值集合，其中
Figure PCTCN2017070764-appb-000001
p(f)为所述候选的推荐特征因子f出现的概率，H(y|f)＝-∑yp(y|f)log(p(y|f))，p(y|f)为y与f的条件概率，如f下推荐信息被点击的概率，及不被点击的概率。
如对于某推荐特征因子f(如性别)，某一广告ad，代表用户点击或者不点击广告的目标值集合y，y可以为1和-1的集合，1可以表示用户点击广告，-1可以表示用户不点击广告，显然也可以设定1表示用户不点击广告，-1表示用户点击广告；本发明实施例可以∑f{p(f)H(y|f)}计算点击率预估模型加入f前，广告的信息熵。
可选的，本发明实施例可根据公式-∑f,ad{p(f,ad)H(y|f,ad)}预估点击率预估模型加入所述候选的推荐特征因子后，推荐信息的第二信息熵；
其中，ad为推荐信息，p(f,ad)为f和ad出现的联合概率，如推荐信息与f同时存在的概率，H(y|f,ad)＝-∑yp(y|f,ad)log(p(y|f,ad))，p(y|f,ad)为y与f及ad的条件概率，如f下针对具体的ad，用户点击该ad的概率及不点击该ad的概率。
如对于某推荐特征因子f，某一广告ad，代表用户点击或者不点击广告的目标值集合y，本发明实施例可以-∑f,ad{p(f,ad)H(y|f,ad)}计算点击率预估模型加入f后，广告ad的信息熵。
步骤S220、根据所述第一信息熵和所述第二信息熵，预估出点击率预估模型加入推荐特征因子前后，推荐信息的信息熵减小量。
本发明实施例可将所述第一信息熵和所述第二信息熵相加，得出所述信息熵减小量；如可根据公式IG＝-∑f,ad{p(f,ad)H(y|f,ad)}+∑f{p(f)H(y|f)}，将所述第一信息熵和所述第二信息熵相结合，其中IG为所述信息熵减小量。
所述信息熵减小量与点击率预估模型加入推荐特征因子前后，推荐信息被点击的增加量相对应。
步骤S230、确定信息熵减小量符合设定信息熵减小条件的推荐特征因子加入所述点击率预估模型。
设定信息熵减小条件为设定量差条件的可选形式，如本发明实施例可将信息熵减小量小于设定信息熵减小值的推荐特征因子，确定为需加入所述点击率预估模型。可选的，本发明实施例也可根据各推荐特征因子的信息熵减小量，对各推荐特征因子进行排序，从而将序位处于设定范围的推荐特征因子，确定为需加入点击率预估模型。
下面以广告推送为例，对图2所示方法的应用例进行说明：
在线广告系统通过算法集合，快速自动的寻找广告、广告环境与受众三者之间的最佳匹配。这种自动化最佳匹配的实现是数据挖掘、信息检索、文本分析、情感计算、机器学习等多种程序算法交互作用的结果。系统从受众进入媒介那一刻，就开始进行着用户模型的数学仿真，借助协同过滤技术确定受众的兴趣模型，利用数据库检索技术寻找最优广告，通过文本分析技术确定广告与投放环境之间的相关度，并结合逻辑回归模型预测出用户点击概率，以对广告进行排序，最终实现广告的精准投放。在点击率预估的过程中，为预估模型选择合适的特征是十分关键的一步；一般而言，点击率预估模型的特征包括三个方面：用户、广告位、广告。
用户侧的特征一般包含用户的年龄、性别、地域、使用上网设备、历史点击及购买行为反应出的兴趣等，用户侧特征一般用于对用户进行精细的分群，学习出用户群对广告的喜好。
广告位侧特征一般包括广告位的位置、大小、上下文等特征，不同广告位上点击率往往存在着天然的差别，广告位上不同的上下文往往也对不同广告的点击率起着相当重要的作用。
用户侧特征的信息增益反应的是不同用户群体点击广告的点击率的差 异，同理，广告位特征的信息增益也是反映了不同的广告位特征对广告点击率的影响。而在本发明实施例中，并不是对于某个人群、某个广告位特征的整体点击率进行预估，而是根据用户特征和广告位特征下具体的推荐特征因子，实现点击率的预估，从而实现对每个广告的点击率进行精准预估。
在此基础上，本发明实施例需确定出哪些推荐特征因子，需加入到点击率预估模型中。本发明可选取一待分析的推荐特征因子f，并针对具体广告ad，设置用户点击或者不点击广告的目标值集合y(y＝1,-1)。
进而，根据公式IG＝-∑f,ad{p(f,ad)H(y|f,ad)}+∑f{p(f)H(y|f)}，预估出点击率预估模型加入推荐特征因子f前后，广告的信息熵减小量，即针对广告ad推荐特征因子f带来的信息增益。对于各候选的推荐特征因子进行类似处理，得到各推荐特征因子对应的信息熵减小量，从而将信息熵减小量符合设定信息熵减小条件的推荐特征因子，确定为需加入点击率预估模型中。
可以看出，本发明实施例是通过推荐特征因子，进行推荐信息点击率的先验预估，从而确定出点击率区分性大的推荐特征因子，实现了较为准确的选择加入点击率预估模型的推荐特征，为提升后续点击率预估的准确性提供了可能。
可选的，在确定需加入点击率预估模型的推荐特征因子后，针对某推荐信息ad的点击行为的后验概率可以下述方式计算：
Figure PCTCN2017070764-appb-000002
进而公式IG＝-∑f,ad{p(f,ad)H(y|f,ad)}+∑f{p(f)H(y|f)}可以进行化简，即
Figure PCTCN2017070764-appb-000003
即
Figure PCTCN2017070764-appb-000004
因此，本发明实施例可根据公式
Figure PCTCN2017070764-appb-000005
预 估出点击率预估模型加入推荐特征因子前后，推荐信息的信息熵减小量；
其中，IG为所述信息熵减小量，f为推荐特征因子，ad为推荐信息，y为推荐信息被点击和不被点击的目标值集合，p(f,ad,y)为f，ad和y出现的联合概率，p(y|f,ad)为y与f及ad的条件概率，p(y|f)为y与f的条件概率。
本发明实施例提供的信息推荐方法，可以较为准确的选择加入点击率预估模型的推荐特征，为提升后续点击率预估的准确性提供可能。
基于前文所述的信息推荐方法，确定出需加入点击率预估模型的推荐特征因子后，本发明实施例可进行信息的推荐，如广告的推荐。
相应的，图3示出了本发明实施例的信息推荐方法中确定向待推荐用户推送的目标推送信息步骤的流程图。参照图3，该步骤可以包括：
步骤S300、调取点击率预估模型及至少一个候选的推荐信息，其中，所述点击率预估模型加入有所确定的推荐特征因子。
所述确定的推荐特征因子，可以在信息推荐时，作为选取点击率预估模型计算所用特征的参考特征；从而根据具体的推荐信息和待推荐用户，从具体的推荐信息和待推荐用户的特征中确定出与该参考特征相应的用户特征因子和信息特征因子，进而通过点击率预估模型，预估出该用户特征因子和信息特征因子相应的点击率。
步骤S310、确定待推荐用户，获取所述待推荐用户的用户特征中，与所述至少一个推荐特征因子相应的用户特征因子。
步骤S320、对于各候选的推荐信息，确定与所述至少一个推荐特征因子相应的信息特征因子。
步骤S330、对于各候选的推荐信息，通过所述点击率预估模型确定与所述用户特征因子及相应的信息特征因子对应的点击率。
步骤S340、根据各候选的推荐信息的点击率，从所述至少一个候选的推荐信息中确定推送给所述待推荐用户的目标推荐信息。
可选的，本发明实施例可按照各候选的推荐信息的点击率，对各候选的推荐信息进行排序，从而将序位处于设定要求的推荐信息，确定为需推送给所述待推荐用户的目标推荐信息。
步骤S350、向所述待推荐用户推送所述目标推荐信息。
以广告的推送为例，图4示出了广告推送的相应示意图。
图4中，广告池用于存放广告主提供的广告素材等。
投放系统用于接受展示页面的请求，根据用户的信息，从广告池拉取符合用户定向的广告，并且参考点击预估模型提供的点击率来进行投放。
点击率预估模型用于当用户有访问广告页面时，可调用点击率预估模型预估用户在该页面上点击广告的概率，该访问广告页面的用户可以认为是待推荐用户；最简单的点击率预估模型是逻辑回归模型，可以由用户u，展示页面d，广告a的特征组合成一个特征向量
Figure PCTCN2017070764-appb-000006
设用户的点击行为y∈(0，1)，逻辑回归可以写为：
Figure PCTCN2017070764-appb-000007
根据用户在历史上的点击行为记录
Figure PCTCN2017070764-appb-000008
可以适用梯度随机下降(SGD)求解出模型参数w；对于新的用户请求，可以通过
Figure PCTCN2017070764-appb-000009
预测该次请求中用户发生点击行为的概率p0。
特征选择用于基于用户点击广告的日志，用户特征等信息，确定出候选的推荐特征因子，以本发明实施例提供的信息推荐方法，分析出对广告的点击率区分性大的推荐特征因子，从而加入到点击率预估模型中，作为点击率预估模型在预估广告点击率时，推荐特征的选取参考。
点击曝光日志用户对广告的点击、曝光行为会以日志的形式记录下来。点击曝光日志的内容例如用户ID，点击或曝光发生的时间等。
用户画像用于通过分析海量用户基础信息数据和点击行为数据，对每个用户赋予特定的标签，以标记用户的兴趣、行为等特征。
本发明实施例通过较为准确的选择加入点击率预估模型的推荐特征，使得点击率预估模型预估出的用户点击推荐信息概率的准确性较高，使得推荐信息的推荐更为精准。
下面对本发明实施例提供的信息推荐装置进行介绍，下文描述的信息推 荐装置可与上文描述的信息推荐方法相互对应参照。
图5为本发明实施例提供的信息推荐装置的结构框图，该装置可应用于具有数据处理能力的电子设备，优选为可进行信息推荐的推荐服务器。参照图5，该信息推荐装置可以包括：
候选因子确定模块100，用于确定至少一个候选的推荐特征因子；
点击信息量差确定模块110，用于预估点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的点击信息量差；
特征选择模块120，用于确定点击信息量差符合设定条件的推荐特征因子，并将所述确定的推荐特征因子加入所述点击率预估模型；以及
推送信息确定模块130，基于所述点击率预估模型确定向待推荐用户推送的目标推送信息。
可选的，图6示出了本发明实施例提供的点击信息量差确定模块110的可选结构，参照图6，点击信息量差确定模块110可以包括：
信息熵确定单元111，用于预估点击率预估模型加入所述候选的推荐特征因子前，推荐信息的第一信息熵，及点击率预估模型加入所述候选的推荐特征因子后，推荐信息的第二信息熵；
信息熵减小确定单元112，用于根据所述第一信息熵和所述第二信息熵，预估出点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的信息熵减小量；所述信息熵减小量与点击率预估模型加入所述候选的推荐特征因子前后，推荐信息被点击的增加量相对应。
可选的，图7示出了本发明实施例提供的信息熵确定单元111的一种可选结构，参照图7，信息熵确定单元111可以包括：
第一信息熵确定子单元1111，用于根据公式∑f{p(f)H(y|f)}预估点击率预估模型加入推荐特征因子前，推荐信息的第一信息熵；
其中，f为推荐特征因子，y为推荐信息被点击和未被点击的目标值集合，其中
Figure PCTCN2017070764-appb-000010
p(f)为推荐特征因子f出现的概率，H(y|f)＝-∑yp(y|f)log(p(y|f))，p(y|f)为y与f的条件概率。
第二信息熵确定子单元1112，用于根据公式-∑f,ad{p(f,ad)H(y|f,ad)}预估 点击率预估模型加入所述候选的推荐特征因子后，推荐信息的第二信息熵；
其中，ad为推荐信息，p(f,ad)为f和ad出现的联合概率，H(y|f,ad)＝-∑yp(y|f,ad)log(p(y|f,ad))，p(y|f,ad)为y与f及ad的条件概率。
相应的，图8示出了信息熵减小确定单元112的可选结构，参照图8，信息熵减小确定单元112可以包括：
信息熵结合子单元1121，用于根据公式IG＝-∑f,ad{p(f,ad)H(y|f,ad)}+∑f{p(f)H(y|f)}，预估推荐信息的信息熵减小量，其中IG为所述信息熵减小量。
可选的，图9示出了本发明实施例提供的点击信息量差确定模块110的另一种可选结构，参照图9，点击信息量差确定模块110可以包括：
计算单元113，用于根据公式
Figure PCTCN2017070764-appb-000011
预估出点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的信息熵减小量；
其中，IG为所述信息熵减小量，f为推荐特征因子，ad为推荐信息，y为推荐信息被点击和未被点击的目标值集合，其中
Figure PCTCN2017070764-appb-000012
p(f,ad,y)为f，ad和y出现的联合概率，p(y|f,ad)为y与f及ad的条件概率，p(y|f)为y与f的条件概率。
本发明实施例提供的信息推荐装置，可以较为准确的选择加入点击率预估模型的推荐特征，为提升后续点击率预估的准确性提供可能。
在本发明的另一实施例中提供了一种信息推荐装置，下文描述的信息推荐装置与上文描述的信息推荐方法相互对应参照。
图10为本发明实施例提供的信息推荐装置中推送信息确定模块130的结构框图。参照图10，该推送信息确定模块130可以包括：
调取单元200，用于调取点击率预估模型及至少一个候选的推荐信息，其中所述点击率预估模型加入所述确定的至少一个推荐特征因子；
用户特征因子确定单元210，用于确定待推荐用户，获取所述待推荐用户 的用户特征中，与所述至少一个推荐特征因子相应的用户特征因子；
信息特征因子确定单元220，用于对于各候选的推荐信息，确定与所述至少一个推荐特征因子相应的信息特征因子；
点击率预估单元230，用于对于各候选的推荐信息，通过所述点击率预估模型确定与所述用户特征因子及相应的信息特征因子对应的点击率；
目标推荐信息确定单元240，用于根据各候选的推荐信息的点击率，从所述至少一个候选的推荐信息中确定推送给所述待推荐用户的目标推荐信息；
推送单元250，用于向所述待推荐用户推送所述目标推荐信息。
可选的，图11示出了装载有信息推荐装置的电子设备的硬件结构框图，该电子设备如推荐服务器，参照图11，该电子设备可以包括：处理器1，通信接口2，存储器3和通信总线4；
其中处理器1、通信接口2、存储器3通过通信总线4完成相互间的通信；
可选的，通信接口2可以为通信模块的接口，如GSM模块的接口；
处理器1，用于执行程序；
存储器3，用于存放程序；
程序可以包括程序代码，所述程序代码包括计算机操作指令。
处理器1可能是一个中央处理器CPU，或者是特定集成电路ASIC(Application Specific Integrated Circuit)，或者是被配置成实施本发明实施例的一个或多个集成电路。
存储器3可能包含高速RAM存储器，也可能还包括非易失性存储器(non-volatile memory)，例如至少一个磁盘存储器。
其中，程序可具体用于：
确定至少一个候选的推荐特征因子；
预估点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的点击信息量差；
确定点击信息量差符合设定条件的推荐特征因子，并将所述确定的推荐特征因子加入所述点击率预估模型；以及
基于所述点击率预估模型确定向待推荐用户推送的目标推送信息。
本说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同相似部分互相参见即可。对于实施例公开的装置而言，由于其与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
专业人员还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
结合本文中所公开的实施例描述的方法或算法的步骤可以直接用硬件、处理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
对所公开的实施例的上述说明，使本领域专业技术人员能够实现或使用本发明。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的，本文中所定义的一般原理可以在不脱离本发明的精神或范围的情况下，在其它实施例中实现。因此，本发明将不会被限制于本文所示的这些实施例，而是要符合与本文所公开的原理和新颖特点相一致的最宽的范围。

Claims (15)

  1. 一种信息推荐方法，其特征在于，包括：
    确定至少一个候选的推荐特征因子；
    预估点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的点击信息量差；
    确定点击信息量差符合设定条件的推荐特征因子，并将所述确定的推荐特征因子加入所述点击率预估模型；以及
    基于所述点击率预估模型确定向待推荐用户推送的目标推送信息。
  2. 根据权利要求1所述的信息推荐方法，其特征在于，所述预估点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的点击信息量差包括：
    预估点击率预估模型加入所述候选的推荐特征因子前，推荐信息的第一信息熵，及点击率预估模型加入所述候选的推荐特征因子后，推荐信息的第二信息熵；以及
    根据所述第一信息熵和所述第二信息熵，预估出点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的信息熵减小量，其中，所述信息熵减小量与点击率预估模型加入推荐特征因子前后，推荐信息被点击的增加量相对应。
  3. 根据权利要求2所述的信息推荐方法，其特征在于，所述对于所述候选的推荐特征因子，预估点击率预估模型加入推荐特征因子前，推荐信息的第一信息熵包括：
    根据公式∑f{p(f)H(y|f)}预估点击率预估模型加入所述候选的推荐特征因子前，推荐信息的第一信息熵；
    其中f为所述候选的推荐特征因子，y为推荐信息被点击和未被点击的目标值集合，其中
    Figure PCTCN2017070764-appb-100001
    p(f)为所述候选的推荐特征因子f出现的概率，H(y|f)＝-∑yp(y|f)log(p(y|f))，p(y|f)为y与f的条件概率。
  4. 根据权利要求3所述的信息推荐方法，其特征在于，所述预估点击率预估模型加入所述候选的推荐特征因子后，推荐信息的第二信息熵包括：
    根据公式-∑f,ad{p(f,ad)H(y|f,ad)}预估点击率预估模型加入所述候选的推荐特征因子后，推荐信息的第二信息熵；
    其中，ad为推荐信息，p(f,ad)为f和ad出现的联合概率，H(y|f,ad)＝-∑yp(y|f,ad)log(p(y|f,ad))，p(y|f,ad)为y与f及ad的条件概率。
  5. 根据权利要求4所述的信息推荐方法，其特征在于，所述根据所述第一信息熵和所述第二信息熵，预估出点击率预估模型加入推荐特征因子前后，推荐信息的信息熵减小量包括：
    根据公式IG＝-∑f,ad{p(f,ad)H(y|f,ad)}+∑f{p(f)H(y|f)}，预估推荐信息的信息熵减小量，其中IG为所述信息熵减小量。
  6. 根据权利要求1所述的信息推荐方法，其特征在于，所述对于所述候选的推荐特征因子，预估点击率预估模型加入推荐特征因子前后，推荐信息的点击信息量差包括：
    根据公式
    Figure PCTCN2017070764-appb-100002
    预估出点击率预估模型加入推荐特征因子前后，推荐信息的信息熵减小量；
    其中，IG为所述信息熵减小量，f为推荐特征因子，ad为推荐信息，y为推荐信息被点击和未被点击的目标值集合，其中
    Figure PCTCN2017070764-appb-100003
    p(f,ad,y)为f，ad和y出现的联合概率，p(y|f,ad)为y与f及ad的条件概率，p(y|f)为y与f的条件概率。
  7. 根据权利要求1-6中任一项所述的信息推荐方法，其特征在于，所述基于所述点击率预估模型确定向待推荐用户推送的目标推送信息包括：
    调取所述点击率预估模型及至少一个候选的推荐信息，其中所述点击率预估模型加入所述确定的推荐特征因子；
    确定待推荐用户，获取所述待推荐用户的用户特征中，与所述至少一个推荐特征因子相应的用户特征因子；
    对于各候选的推荐信息，确定与所述至少一个推荐特征因子相应的信息特征因子；
    对于各候选的推荐信息，通过所述点击率预估模型确定与所述用户特征因子及相应的信息特征因子对应的点击率；
    根据各候选的推荐信息的点击率，从所述至少一个候选的推荐信息中确定推送给所述待推荐用户的目标推荐信息；以及
    向所述待推荐用户推送所述目标推荐信息。
  8. 一种信息推荐装置，其特征在于，包括：
    候选因子确定模块，用于确定至少一个候选的推荐特征因子；
    点击信息量差确定模块，用于预估点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的点击信息量差；
    特征选择模块，用于确定点击信息量差符合设定条件的推荐特征因子，并将所述确定的推荐特征因子加入所述点击率预估模型；以及
    推送信息确定模块，基于所述点击率预估模型确定向待推荐用户推送的目标推送信息。
  9. 根据权利要求8所述的信息推荐装置，其特征在于，所述点击信息量差确定模块包括：
    信息熵确定单元，用于预估点击率预估模型加入所述候选的推荐特征因子前，推荐信息的第一信息熵，及点击率预估模型加入所述候选的推荐特征因子后，推荐信息的第二信息熵；
    信息熵减小确定单元，用于根据所述第一信息熵和所述第二信息熵，预估出点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的信息熵减小量；其中，所述信息熵减小量与点击率预估模型加入所述候选的推荐特征因子前后，推荐信息被点击的增加量相对应。
  10. 根据权利要求9所述的信息推荐装置，其特征在于，所述信息熵确定单元包括：
    第一信息熵确定子单元，用于根据公式∑f{p(f)H(y|f)}预估点击率预估模型加入推荐特征因子前，推荐信息的第一信息熵；
    其中f为推荐特征因子，y为推荐信息被点击和未被点击的目标值集合，其中
    Figure PCTCN2017070764-appb-100004
    p(f)为推荐特征因子f出现的概率， H(y|f)＝-∑yp(y|f)log(p(y|f))，p(y|f)为y与f的条件概率。
  11. 根据权利要求10所述的信息推荐装置，其特征在于，所述信息熵确定单元还包括：
    第二信息熵确定子单元，用于根据公式-∑f,ad{p(f,ad)H(y|f,ad)}预估点击率预估模型加入推荐特征因子后，推荐信息的第二信息熵；
    其中，ad为推荐信息，p(f,ad)为f和ad出现的联合概率，H(y|f,ad)＝-∑yp(y|f,ad)log(p(y|f,ad))，p(y|f,ad)为y与f及ad的条件概率。
  12. 根据权利要求11所述的信息推荐装置，其特征在于，所述信息熵减小确定单元包括：
    信息熵结合子单元，用于根据公式IG＝-∑f,ad{p(f,ad)H(y|f,ad)}+∑f{p(f)H(y|f)}，预估推荐信息的信息熵减小量，其中IG为所述信息熵减小量。
  13. 根据权利要求8所述的信息推荐装置，其特征在于，所述点击信息量差确定模块包括：
    计算单元，用于根据公式
    Figure PCTCN2017070764-appb-100005
    预估出点击率预估模型加入所述候选的推荐特征因子前后，推荐信息的信息熵减小量；
    其中，IG为所述信息熵减小量，f为推荐特征因子，ad为推荐信息，y为推荐信息被点击和未被点击的目标值集合，其中
    Figure PCTCN2017070764-appb-100006
    p(f,ad,y)为f，ad和y出现的联合概率，p(y|f,ad)为y与f及ad的条件概率，p(y|f)为y与f的条件概率。
  14. 根据权利要求8-13中任一项所述的信息推荐装置，其特征在于，所述推送信息确定模块包括：
    调取单元，用于调取所述点击率预估模型及至少一个候选的推荐信息，其中，所述点击率预估模型加入所述确定的至少一个推荐特征因子；
    用户特征因子确定单元，用于确定待推荐用户，获取所述待推荐用户的用户特征中，与所述至少一个推荐特征因子相应的用户特征因子；
    信息特征因子确定单元，用于对于各候选的推荐信息，确定与所述至少一个推荐特征因子相应的信息特征因子；
    点击率预估单元，用于对于各候选的推荐信息，通过所述点击率预估模型确定与所述用户特征因子及相应的信息特征因子对应的点击率；
    目标推荐信息确定单元，用于根据各候选的推荐信息的点击率，从所述至少一个候选的推荐信息中确定推送给所述待推荐用户的目标推荐信息；以及
    推送单元，用于向所述待推荐用户推送所述目标推荐信息。
  15. 一种非易失性存储介质，用于存储一个或多个计算机程序，其中，所述计算机程序包括具有一个或多个存储器的处理器可运行的指令；所述处理器运行该指令以执行根据权利要求1-7中任一项所述的信息推荐方法。
PCT/CN2017/070764 2016-01-12 2017-01-10 信息推荐方法及装置 WO2017121314A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
KR1020177034700A KR101999471B1 (ko) 2016-01-12 2017-01-10 정보 추천 방법 및 장치
JP2017562719A JP6588572B2 (ja) 2016-01-12 2017-01-10 情報推薦方法および情報推薦装置
US15/825,396 US11301525B2 (en) 2016-01-12 2017-11-29 Method and apparatus for processing information

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610017754.7 2016-01-12
CN201610017754.7A CN105678587B (zh) 2016-01-12 2016-01-12 一种推荐特征确定方法、信息推荐方法及装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15/825,396 Continuation US11301525B2 (en) 2016-01-12 2017-11-29 Method and apparatus for processing information

Publications (1)

Publication Number Publication Date
WO2017121314A1 true WO2017121314A1 (zh) 2017-07-20

Family

ID=56300069

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/070764 WO2017121314A1 (zh) 2016-01-12 2017-01-10 信息推荐方法及装置

Country Status (5)

Country Link
US (1) US11301525B2 (zh)
JP (1) JP6588572B2 (zh)
KR (1) KR101999471B1 (zh)
CN (1) CN105678587B (zh)
WO (1) WO2017121314A1 (zh)

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105678587B (zh) 2016-01-12 2020-11-24 腾讯科技（深圳）有限公司 一种推荐特征确定方法、信息推荐方法及装置
CN108268464B (zh) * 2016-12-30 2021-01-12 广东精点数据科技股份有限公司 一种基于协同过滤与logistic回归的个性化推荐方法及装置
CN107172151B (zh) * 2017-05-18 2020-08-07 百度在线网络技术（北京）有限公司 用于推送信息的方法和装置
CN107613022B (zh) * 2017-10-20 2020-10-16 阿里巴巴（中国）有限公司 内容推送方法、装置及计算机设备
US11238358B2 (en) * 2017-12-18 2022-02-01 Microsoft Technology Licensing, Llc Predicting site visit based on intervention
CN110110210A (zh) * 2018-01-22 2019-08-09 腾讯科技（北京）有限公司 推送展示信息的方法和装置
JP7217096B2 (ja) * 2018-06-26 2023-02-02 株式会社Lifull 広告オプション割当装置、広告オプション割当プログラム、広告オプション割当方法、及び情報提供システム
JP7217097B2 (ja) * 2018-06-26 2023-02-02 株式会社Lifull 照会確率提示装置、照会確率提示プログラム、照会確率提示方法、及び情報提供システム
CN110781374A (zh) * 2018-07-13 2020-02-11 北京字节跳动网络技术有限公司 用户数据匹配方法、装置、电子设备和计算机可读介质
CN109255070B (zh) * 2018-08-01 2022-04-12 百度在线网络技术（北京）有限公司 推荐信息处理方法、装置、计算机设备及存储介质
CN109165742A (zh) * 2018-08-28 2019-01-08 北京百度网讯科技有限公司 推荐方法、装置、存储介质和终端设备
CN109272360B (zh) * 2018-09-28 2021-09-10 有米科技股份有限公司 一种广告智能推荐方法、系统及装置
CN111598638B (zh) * 2019-02-21 2023-11-07 北京沃东天骏信息技术有限公司 点击率确定方法、装置及设备
CN110413358B (zh) * 2019-06-27 2022-03-04 创新先进技术有限公司 页面的展示方法及装置
CN111159241B (zh) * 2019-12-20 2023-04-07 深圳前海微众银行股份有限公司 一种点击转化预估方法及装置
CN111914172B (zh) * 2020-07-29 2021-09-10 上海梅斯医药科技有限公司 一种基于用户标签的医学信息推荐方法及系统
CN112115365B (zh) * 2020-09-25 2021-09-14 贝壳找房(北京)科技有限公司 模型协同优化的方法、装置、介质和电子设备
CN113763103B (zh) * 2021-01-19 2024-06-14 北京沃东天骏信息技术有限公司 一种物品搭配关系确定方法、装置、电子设备和存储介质
US11803253B2 (en) * 2021-11-29 2023-10-31 International Business Machines Corporation Keyword recommendations for virtual keyboards

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103092877A (zh) * 2011-11-04 2013-05-08 百度在线网络技术（北京）有限公司 一种关键词推荐方法和装置
US20130275580A1 (en) * 2012-04-11 2013-10-17 Taboola.Com Ltd. Click through rate estimation in varying display situations
CN104572734A (zh) * 2013-10-23 2015-04-29 腾讯科技（深圳）有限公司 问题推荐方法、装置及系统
CN105046277A (zh) * 2015-07-15 2015-11-11 华南农业大学 特征显著性在图像质量评价中的鲁棒机理研究方法
CN105678587A (zh) * 2016-01-12 2016-06-15 腾讯科技（深圳）有限公司 一种推荐特征确定方法、信息推荐方法及装置

Family Cites Families (88)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6460036B1 (en) * 1994-11-29 2002-10-01 Pinpoint Incorporated System and method for providing customized electronic newspapers and target advertisements
US5948061A (en) * 1996-10-29 1999-09-07 Double Click, Inc. Method of delivery, targeting, and measuring advertising over networks
US6286005B1 (en) * 1998-03-11 2001-09-04 Cannon Holdings, L.L.C. Method and apparatus for analyzing data and advertising optimization
US6907566B1 (en) * 1999-04-02 2005-06-14 Overture Services, Inc. Method and system for optimum placement of advertisements on a webpage
US6981040B1 (en) * 1999-12-28 2005-12-27 Utopy, Inc. Automatic, personalized online information and product services
US20040151250A1 (en) * 2001-06-11 2004-08-05 Witterbrood Rimmert B. Feature point selection
EP1896958A4 (en) * 2005-06-28 2010-08-18 Choicestream Inc METHODS AND APPARATUS FOR STATISTICAL ADVERTISING ADVERTISEMENT TARGETING SYSTEM
US20100138451A1 (en) * 2006-04-03 2010-06-03 Assaf Henkin Techniques for facilitating on-line contextual analysis and advertising
US7937345B2 (en) * 2006-07-12 2011-05-03 Kofax, Inc. Data classification methods using machine learning techniques
US20080103897A1 (en) * 2006-10-25 2008-05-01 Microsoft Corporation Normalizing and tracking user attributes for transactions in an advertising exchange
JP5312771B2 (ja) * 2006-10-26 2013-10-09 株式会社エム・シー・エヌ クエリに応答して、関連性のある広告を決定する技術
US8175914B1 (en) * 2007-07-30 2012-05-08 Google Inc. Automatic adjustment of advertiser bids to equalize cost-per-conversion among publishers for an advertisement
US20090063268A1 (en) * 2007-09-04 2009-03-05 Burgess David A Targeting Using Historical Data
US7912843B2 (en) * 2007-10-29 2011-03-22 Yahoo! Inc. Method for selecting electronic advertisements using machine translation techniques
US9203912B2 (en) * 2007-11-14 2015-12-01 Qualcomm Incorporated Method and system for message value calculation in a mobile environment
US9734460B1 (en) * 2008-06-13 2017-08-15 Google Inc. Adjusting participation of content in a selection process
US8738436B2 (en) * 2008-09-30 2014-05-27 Yahoo! Inc. Click through rate prediction system and method
US20100262456A1 (en) * 2009-04-08 2010-10-14 Jun Feng System and Method for Deep Targeting Advertisement Based on Social Behaviors
JP5519210B2 (ja) * 2009-08-10 2014-06-11 株式会社エイチアイ モデル生成装置、モデル生成方法、情報処理システム、及びプログラム
US20110093331A1 (en) * 2009-10-19 2011-04-21 Donald Metzler Term Weighting for Contextual Advertising
US9760802B2 (en) * 2010-01-27 2017-09-12 Ebay Inc. Probabilistic recommendation of an item
US20110313842A1 (en) * 2010-03-24 2011-12-22 Taykey Ltd. System and methods thereof for providing an advertisement placement recommendation based on trends
US8655695B1 (en) * 2010-05-07 2014-02-18 Aol Advertising Inc. Systems and methods for generating expanded user segments
US8510658B2 (en) * 2010-08-11 2013-08-13 Apple Inc. Population segmentation
US20120150626A1 (en) * 2010-12-10 2012-06-14 Zhang Ruofei Bruce System and Method for Automated Recommendation of Advertisement Targeting Attributes
US9104960B2 (en) * 2011-06-20 2015-08-11 Microsoft Technology Licensing, Llc Click prediction using bin counting
WO2013052936A1 (en) * 2011-10-06 2013-04-11 Infersystems Corp. Automated allocation of media via network
CN102346899A (zh) * 2011-10-08 2012-02-08 亿赞普(北京)科技有限公司 一种基于用户行为的广告点击率预测方法和装置
US20130138507A1 (en) * 2011-11-30 2013-05-30 Amit Kumar Predictive modeling for e-commerce advertising systems and methods
US20130191223A1 (en) * 2012-01-20 2013-07-25 Visa International Service Association Systems and methods to determine user preferences for targeted offers
US20140372350A1 (en) * 2012-02-09 2014-12-18 Kenshoo Ltd. System, A Method and a Computer Program Product for Performance Assessment
US20130211905A1 (en) * 2012-02-13 2013-08-15 Microsoft Corporation Attractiveness-based online advertisement click prediction
CN103246686A (zh) * 2012-02-14 2013-08-14 阿里巴巴集团控股有限公司 文本分类方法和装置及文本分类的特征处理方法和装置
JP2013205170A (ja) * 2012-03-28 2013-10-07 Sony Corp 情報処理装置、情報処理方法、およびプログラム
US20130346182A1 (en) * 2012-06-20 2013-12-26 Yahoo! Inc. Multimedia features for click prediction of new advertisements
CN102821465B (zh) * 2012-09-07 2014-11-26 哈尔滨工业大学 基于分区信息熵增益的wlan室内定位方法
US20140101685A1 (en) * 2012-10-04 2014-04-10 Lucid Commerce, Inc. Continuous optimization of advertising campaigns based on real time tracking
US20140149223A1 (en) * 2012-11-29 2014-05-29 Nipun Mathur Targeted Advertisements In Mobile Applications
US9129227B1 (en) * 2012-12-31 2015-09-08 Google Inc. Methods, systems, and media for recommending content items based on topics
US20140207564A1 (en) * 2013-01-18 2014-07-24 Linkedln Corporation System and method for serving electronic content
US20140236705A1 (en) * 2013-02-15 2014-08-21 Turn Inc. Method and apparatus for data-driven multi-touch attribution determination in multichannel advertising campaigns
US20140372202A1 (en) * 2013-06-17 2014-12-18 Google Inc. Predicting performance of content items using loss functions
US20150006286A1 (en) * 2013-06-28 2015-01-01 Linkedin Corporation Targeting users based on categorical content interactions
US20150006295A1 (en) * 2013-06-28 2015-01-01 Linkedln Corporation Targeting users based on previous advertising campaigns
US20150006294A1 (en) * 2013-06-28 2015-01-01 Linkedln Corporation Targeting rules based on previous recommendations
US9767489B1 (en) * 2013-08-30 2017-09-19 Google Inc. Content item impression effect decay
US20160255139A1 (en) * 2016-03-12 2016-09-01 Yogesh Chunilal Rathod Structured updated status, requests, user data & programming based presenting & accessing of connections or connectable users or entities and/or link(s)
US10832281B1 (en) * 2013-12-20 2020-11-10 Groupon, Inc. Systems, apparatus, and methods for providing promotions based on consumer interactions
WO2015096146A1 (en) * 2013-12-27 2015-07-02 Yahoo! Inc. Systems and methods for a unified audience targeting solution
CN103747537B (zh) * 2014-01-15 2017-05-03 广东交通职业技术学院 一种基于熵度量的无线传感器网络离群数据自适应检测方法
CN104834641B (zh) * 2014-02-11 2019-03-15 腾讯科技（北京）有限公司 网络媒介信息的处理方法和相关系统
US20150235260A1 (en) * 2014-02-20 2015-08-20 Linkedln Corporation Forecasting electronic events
US9727818B1 (en) * 2014-02-23 2017-08-08 Google Inc. Impression effect modeling for content items
EP4219071A3 (en) * 2014-03-13 2023-08-09 The Nielsen Company (US), LLC Methods and apparatus to compensate impression data for misattribution and/or non-coverage by a database proprietor
US9836765B2 (en) * 2014-05-19 2017-12-05 Kibo Software, Inc. System and method for context-aware recommendation through user activity change detection
US9886670B2 (en) * 2014-06-30 2018-02-06 Amazon Technologies, Inc. Feature processing recipes for machine learning
US10452992B2 (en) * 2014-06-30 2019-10-22 Amazon Technologies, Inc. Interactive interfaces for machine learning model evaluations
US20150379571A1 (en) * 2014-06-30 2015-12-31 Yahoo! Inc. Systems and methods for search retargeting using directed distributed query word representations
US10963810B2 (en) * 2014-06-30 2021-03-30 Amazon Technologies, Inc. Efficient duplicate detection for machine learning data sets
US10528981B2 (en) * 2014-07-18 2020-01-07 Facebook, Inc. Expansion of targeting criteria using an advertisement performance metric to maintain revenue
US10318983B2 (en) * 2014-07-18 2019-06-11 Facebook, Inc. Expansion of targeting criteria based on advertisement performance
US10706439B1 (en) * 2014-07-30 2020-07-07 Groupon, Inc. Predictive recommendation system using tiered feature data
US20160180248A1 (en) * 2014-08-21 2016-06-23 Peder Regan Context based learning
US20160071162A1 (en) * 2014-09-10 2016-03-10 Sysomos L.P. Systems and Methods for Continuous Analysis and Procurement of Advertisement Campaigns
JP6129802B2 (ja) * 2014-09-19 2017-05-17 ヤフー株式会社 情報処理装置、情報処理方法、および情報処理プログラム
US10296546B2 (en) * 2014-11-24 2019-05-21 Adobe Inc. Automatic aggregation of online user profiles
US20160189239A1 (en) * 2014-12-30 2016-06-30 Yahoo!, Inc. Advertisement generator
US10671679B2 (en) * 2014-12-30 2020-06-02 Oath Inc. Method and system for enhanced content recommendation
US20160210657A1 (en) * 2014-12-30 2016-07-21 Anto Chittilappilly Real-time marketing campaign stimuli selection based on user response predictions
US20160253325A1 (en) * 2014-12-30 2016-09-01 Socialtopias, Llc Method and apparatus for programmatically adjusting the relative importance of content data as behavioral data changes
US20160189202A1 (en) * 2014-12-31 2016-06-30 Yahoo! Inc. Systems and methods for measuring complex online strategy effectiveness
US20160210658A1 (en) * 2014-12-31 2016-07-21 Anto Chittilappilly Determining touchpoint attributions in a segmented media campaign
US20160232575A1 (en) * 2015-02-06 2016-08-11 Facebook, Inc. Determining a number of cluster groups associated with content identifying users eligible to receive the content
US10497024B2 (en) * 2015-04-28 2019-12-03 Facebook, Inc. Identifying content to present to a group of online system users based on user actions and specified by a third-party system
WO2016201631A1 (en) * 2015-06-17 2016-12-22 Yahoo! Inc. Systems and methods for online content recommendation
CN105183772A (zh) * 2015-08-07 2015-12-23 百度在线网络技术（北京）有限公司 投放信息点击率预估方法及装置
US10127573B2 (en) * 2015-11-13 2018-11-13 Facebook, Inc. Framework for evaluating targeting models
US10057199B2 (en) * 2015-11-16 2018-08-21 Facebook, Inc. Ranking and filtering comments based on impression calculations
US10242386B2 (en) * 2015-12-16 2019-03-26 Facebook, Inc. Grouping users into tiers based on similarity to a group of seed users
US20170178199A1 (en) * 2015-12-22 2017-06-22 Intuit Inc. Method and system for adaptively providing personalized marketing experiences to potential customers and users of a tax return preparation system
US20170186047A1 (en) * 2015-12-28 2017-06-29 International Business Machines Corporation Optimization of audience groups in online advertising bidding
US20170236073A1 (en) * 2016-02-12 2017-08-17 Linkedln Corporation Machine learned candidate selection on inverted indices
US10200485B2 (en) * 2016-04-05 2019-02-05 Facebook, Inc. Pushing news feed content to client devices
US20180189074A1 (en) * 2017-01-03 2018-07-05 Facebook, Inc. Postview of content items for dynamic creative optimization
US20190197176A1 (en) * 2017-12-21 2019-06-27 Microsoft Technology Licensing, Llc Identifying relationships between entities using machine learning
US10740825B1 (en) * 2018-04-10 2020-08-11 Facebook, Inc. User clustering in a latent space for identifying user interest in a content item
CN109614794A (zh) * 2018-12-12 2019-04-12 中国科学院计算机网络信息中心 基于信息增益的特征选择方法、装置及存储介质
CN111159564A (zh) * 2019-12-31 2020-05-15 联想(北京)有限公司 信息推荐方法、装置、存储介质及计算机设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103092877A (zh) * 2011-11-04 2013-05-08 百度在线网络技术（北京）有限公司 一种关键词推荐方法和装置
US20130275580A1 (en) * 2012-04-11 2013-10-17 Taboola.Com Ltd. Click through rate estimation in varying display situations
CN104572734A (zh) * 2013-10-23 2015-04-29 腾讯科技（深圳）有限公司 问题推荐方法、装置及系统
CN105046277A (zh) * 2015-07-15 2015-11-11 华南农业大学 特征显著性在图像质量评价中的鲁棒机理研究方法
CN105678587A (zh) * 2016-01-12 2016-06-15 腾讯科技（深圳）有限公司 一种推荐特征确定方法、信息推荐方法及装置

Also Published As

Publication number Publication date
JP6588572B2 (ja) 2019-10-09
CN105678587B (zh) 2020-11-24
KR101999471B1 (ko) 2019-10-01
CN105678587A (zh) 2016-06-15
US11301525B2 (en) 2022-04-12
US20180081978A1 (en) 2018-03-22
KR20170141247A (ko) 2017-12-22
JP2018526710A (ja) 2018-09-13

Similar Documents

Publication Publication Date Title
WO2017121314A1 (zh) 信息推荐方法及装置
WO2020140400A1 (zh) 基于用户行为的产品推荐方法、装置、设备及存储介质
US8738436B2 (en) Click through rate prediction system and method
US10348550B2 (en) Method and system for processing network media information
US8311957B2 (en) Method and system for developing a classification tool
WO2017101240A1 (zh) 一种信息处理方法及服务器、计算机存储介质
WO2017190610A1 (zh) 目标用户定向方法、装置和计算机存储介质
WO2018053966A1 (zh) 点击率预估
US11593860B2 (en) Method, medium, and system for utilizing item-level importance sampling models for digital content selection policies
US8600809B1 (en) Predictive model performance
US11107133B2 (en) Systems and methods for self-serve marketing pages with multi-armed bandit
US20100228558A1 (en) Aggregate Content-Based Advertising
US20210056458A1 (en) Predicting a persona class based on overlap-agnostic machine learning models for distributing persona-based digital content
US11288709B2 (en) Training and utilizing multi-phase learning models to provide digital content to client devices in a real-time digital bidding environment
JP2008547129A (ja) 自動広告掲載
WO2016107354A1 (zh) 提供用户个性化资源消息推送的方法和装置
WO2019061664A1 (zh) 电子装置、基于用户上网数据的产品推荐方法及存储介质
US20180285748A1 (en) Performance metric prediction for delivery of electronic media content items
WO2023000491A1 (zh) 一种应用推荐方法、装置、设备及计算机可读存储介质
JP5061999B2 (ja) 解析装置、解析方法及び解析プログラム
Vasconcelos et al. What makes your opinion popular? Predicting the popularity of micro-reviews in Foursquare
US20160055203A1 (en) Method for record selection to avoid negatively impacting latency
CN116109354A (zh) 内容推荐方法、装置、设备、存储介质及计算机程序产品
CN110457464B (zh) 用于信息处理的方法、装置和计算设备
US20150332192A1 (en) Estimating Future Inventory

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17738128

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 20177034700

Country of ref document: KR

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2017562719

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17738128

Country of ref document: EP

Kind code of ref document: A1