Information Recommendation Method, Apparatus and Server

Abstract

The present invention provides an information recommendation method, apparatus and server, relating to the field of computer technology. The information recommendation method includes: determining preference tags of a target user from multiple tags of a historical video according to the operation behavior of the target user on the historical video, where the historical video is a video that the target user has watched historically; determining multiple interested themes of the target user from multiple themes corresponding to multiple theme tags according to the multiple theme tags in a theme library and the preference tags; determining a target theme from the multiple interested themes according to the video content in the multiple interested themes; and recommending the video content in the target theme to a terminal device corresponding to the target user. By automatically determining the preference tags of the target user and the multiple interested themes of the target user, and then determining the target theme, it is possible to automatically determine the target theme without manual screening, saving human resources and improving the efficiency of theme recommendation.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular, to an information recommendation method, apparatus, and server. Background Art

[0002] Internet TV refers to the transmission of video, audio, graphics, text, data, etc. based on the public Internet, and through terminal receiving devices such as televisions and set-top boxes, to provide the public with safe, reliable, interactive, and manageable multimedia audio-visual services. With the popularization of Internet TV, theme recommendation is also widely used in TV products.

[0003] In the related art, the theme recommendation method is that the operation personnel define and generate the core concept of the theme based on operation experience and popular trends, and then manually screen out the programs that conform to the defined core concept from the defined core concept to form a theme, and deploy the theme to the terminal to achieve theme recommendation.

[0004] However, in the related art, it is necessary to manually screen out programs to form a theme, which wastes unnecessary human resources and reduces the efficiency of theme recommendation. Summary of the Invention

[0005] The purpose of the present invention is to provide an information recommendation method, apparatus, server, and storage medium for solving the above problems existing in the related art in view of the above deficiencies in the prior art.

[0006] To achieve the above object, the technical solutions adopted in the embodiments of the present invention are as follows:

[0007] In a first aspect, an embodiment of the present invention provides an information recommendation method, including:

[0008] Determining preference tags of a target user from multiple tags of the historical video according to an operation behavior of the target user on the historical video, where the historical video is a video that the target user has watched historically;

[0009] Determining multiple interested themes of the target user from multiple themes corresponding to the multiple theme tags according to the multiple theme tags in a theme library and the preference tags;

[0010] Determining a target theme from the multiple interested themes according to video content in the multiple interested themes;

[0011] Recommending video content in the target theme to a terminal device corresponding to the target user.

[0012] Optionally, the determining preference tags of the target user from multiple tags of the historical video according to an operation behavior of the target user on the historical video includes:

[0013] Determine the preference scores of the target user for multiple tags in each tag dimension according to the operation behaviors of the target user on historical videos;

[0014] Sort the multiple tags in each tag dimension according to the preference scores of the multiple tags in each tag dimension to obtain the sorted multiple tags in each tag dimension;

[0015] Determine the preferred tags in each tag dimension from the sorted multiple tags in each tag dimension.

[0016] Optionally, the determining the preference scores of the target user for multiple tags in each tag dimension according to the operation behaviors of the target user on historical videos includes:

[0017] For each historical time period, according to the behavior type of the operation behavior, the behavior weight of the behavior type, the number of times the target user performs the operation behavior on each tag in the multiple tags in each tag dimension of the target user, the number of times the target user performs the operation behavior on all tags under the target tag dimension corresponding to each tag, the number of times all users perform the operation behavior on all tags under the target tag dimension, and the number of times all users perform the operation behavior on each tag, obtain the initial score of the target user for each tag in each historical time period;

[0018] Accumulate the initial scores for each historical time period to obtain the preference scores of the target user for multiple tags in each tag dimension.

[0019] Optionally, the determining the multiple interested topics of the target user from the multiple topics corresponding to the multiple topic tags according to the multiple topic tags in the topic library and the preferred tags includes:

[0020] Match the multiple topic tags and the preferred tags to obtain the interest scores of the target user for the multiple topics;

[0021] Sort the multiple topics according to the interest scores of the multiple topics to obtain the sorted multiple topics;

[0022] Determine the multiple interested topics of the target user from the sorted multiple topics.

[0023] Optionally, the matching the multiple topic tags and the preferred tags to obtain the interest scores of the target user for the multiple topics includes:

[0024] Calculate the interest scores of the multiple topics based on the preset weights corresponding to the tag dimensions, the number of matches between the preference tags and each topic tag on the corresponding tag dimensions, and the preset values in the tag dimensions.

[0025] Optionally, determining a target topic from the multiple interesting topics according to the video content in the multiple interesting topics includes:

[0026] Rate the video content in each interesting topic to obtain the scores of each video content in each interesting topic;

[0027] Determine the target video content of each interesting topic according to the scores of each video content in each interesting topic;

[0028] Obtain the scores of each interesting topic according to the scores of the target video content of each interesting topic;

[0029] Determine the target topic from the multiple interesting topics according to the scores of each interesting topic.

[0030] Optionally, rating the video content in each interesting topic to obtain the scores of each video content in each interesting topic includes:

[0031] Obtain the feature information of the target user and the feature information of the video content in each interesting topic;

[0032] Use a preset video content rating model to process the feature information of the target user and the feature information of the video content in each interesting topic to obtain the scores of each video content in each interesting topic.

[0033] Optionally, the method further includes:

[0034] Train a model according to the feature information of the sample user, the first sample video content that is exposed and has been clicked by the sample user, and the second sample video content that is exposed and has not been clicked by the sample user to obtain the preset video content rating model.

[0035] In a second aspect, an information recommendation device provided by an embodiment of the present invention further includes:

[0036] A determination module, configured to determine preference tags of the target user from multiple tags of the historical video according to the operation behavior of the target user on the historical video, where the historical video is a video that the target user has watched historically; determine multiple interested themes of the target user from multiple themes corresponding to the multiple theme tags according to the multiple theme tags in the theme library and the preference tags; determine a target theme from the multiple interested themes according to the video content in the multiple interested themes;

[0037] A recommendation module, configured to recommend video content in the target theme to a terminal device corresponding to the target user.

[0038] Optionally, the determination module is specifically configured to determine preference scores of the target user for multiple tags in each tag dimension according to the operation behavior of the target user on the historical video; sort the multiple tags in each tag dimension according to the preference scores of the multiple tags in each tag dimension to obtain the sorted multiple tags in each tag dimension; determine the preference tags in each tag dimension from the sorted multiple tags in each tag dimension.

[0039] Optionally, the determination module is specifically configured to, for each historical time period, obtain an initial score of the target user for each tag according to the behavior type of the operation behavior, the behavior weight of the behavior type, the number of times the target user performs the operation behavior on each tag in the multiple tags in each tag dimension, the number of times the target user performs the operation behavior on all tags in the target tag dimension corresponding to each tag, the number of times all users perform the operation behavior on all tags in the target tag dimension, and the number of times all users perform the operation behavior on each tag; accumulate the initial scores for each historical time period to obtain preference scores of the target user for multiple tags in each tag dimension.

[0040] Optionally, the determination module is specifically configured to match the multiple theme tags and the preference tags to obtain interest scores of the target user for the multiple themes; sort the multiple themes according to the interest scores of the multiple themes to obtain the sorted multiple themes; determine multiple interested themes of the target user from the sorted multiple themes.

[0041] Optionally, the determination module is specifically configured to calculate interest scores of the multiple themes according to a preset weight corresponding to the tag dimension, the number of matches between the preference tag and each theme tag in the corresponding tag dimension, and a preset value in the tag dimension.

[0042] Optionally, the determining module is specifically configured to score the video content in each topic of interest to obtain the scores of each video content in each topic of interest; determine the target video content of each topic of interest according to the scores of each video content in each topic of interest; obtain the scores of each topic of interest according to the scores of the target video content of each topic of interest; and determine the target topic from the multiple topics of interest according to the scores of each topic of interest.

[0043] Optionally, the determining module is specifically configured to obtain the feature information of the target user and the feature information of the video content in each topic of interest; and use a preset video content scoring model to process the feature information of the target user and the feature information of the video content in each topic of interest to obtain the scores of each video content in each topic of interest.

[0044] Optionally, the apparatus further includes:

[0045] A training module, configured to perform model training according to the feature information of a sample user, a first sample video content that has been exposed and played by the sample user, and a second sample video content that has been exposed but not played by the sample user, to obtain the preset video content scoring model.

[0046] In a third aspect, an embodiment of the present invention further provides a server, including: a memory and a processor, where the memory stores a computer program executable by the processor, and when the processor executes the computer program, the information recommendation method according to any one of the above first aspects is implemented.

[0047] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the storage medium, and when the computer program is read and executed, the information recommendation method according to any one of the above first aspects is implemented.

[0048] The beneficial effects of the present invention are as follows: An information recommendation method provided by an embodiment of the present invention includes: determining preference tags of a target user from multiple tags of historical videos according to the operation behaviors of the target user for the historical videos, where the historical videos are the videos watched by the target user in history; determining multiple interested themes of the target user from multiple themes corresponding to multiple theme tags in a theme library according to the multiple theme tags and the preference tags; determining a target theme from the multiple interested themes according to the video content in the multiple interested themes; and recommending the video content in the target theme to the terminal device corresponding to the target user. By automatically determining the preference tags of the target user and the multiple interested themes of the target user, and then determining the target theme from the multiple interested themes according to the video content in the multiple interested themes, it is possible to automatically determine the target theme without manual screening, saving human resources and improving the efficiency of theme recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0050] Figure 1 It is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention;

[0051] Figure 2 It is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention;

[0052] Figure 3 It is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention;

[0053] Figure 4 It is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention;

[0054] Figure 5 It is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention;

[0055] Figure 6 It is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention;

[0056] Figure 7 It is a schematic structural diagram of an information recommendation device provided by an embodiment of the present invention;

[0057] Figure 8 It is a schematic structural diagram of a server provided by an embodiment of the present invention. Detailed implementation mode

[0058] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention.

[0059] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application claimed, but merely represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the scope of protection of the present application.

[0060] In the description of the present application, it should be noted that if terms such as "upper", "lower", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of this application is usually placed during use, it is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present application.

[0061] In addition, terms such as "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0062] It should be noted that, without conflict, the features in the embodiments of the present application can be combined with each other.

[0063] In related technologies, the topic recommendation method is that operators define and generate the core concepts of topics based on operation experience and trends, and then manually screen out programs that meet the defined core concepts to form topics, and deploy the topics to the terminals to achieve topic recommendation. However, in related technologies, it is necessary to manually screen out programs to form topics, which wastes unnecessary human resources and reduces the efficiency of topic recommendation.

[0064] In view of the above technical problems existing in the related art, an information recommendation method provided by an embodiment of the present application determines preference tags of a target user, and determines multiple interested topics of the target user from multiple topics corresponding to multiple topic tags in a topic library according to the multiple topic tags and the preference tags. Then, according to the video content in the multiple interested topics, a target topic, that is, the topic to be recommended, is determined from the multiple interested topics, and the video content in the target topic is recommended to the terminal device corresponding to the target user; it can automatically determine the target topic without manual screening, saving human resources and improving the efficiency of topic recommendation.

[0065] The information recommendation method provided by an embodiment of the present application can be applied to a server. The following explains the information recommendation method provided by an embodiment of the present application.

[0066] Figure 1 It is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention, as Figure 1 shown. The method may include:

[0067] S101. Determine preference tags of the target user from multiple tags of historical videos according to the operation behaviors of the target user for historical videos.

[0068] Among them, the historical videos are the videos that the target user has watched historically. The operation behaviors may include: search behavior, play behavior, click behavior.

[0069] In some embodiments, according to the operation behaviors of the target user for historical videos, preference tags of the target user for each tag dimension are determined from multiple tags of multiple tag dimensions of historical videos, where each tag dimension of historical videos has multiple tags.

[0070] In practical applications, the multiple tag dimensions may include: first-level classification, second-level classification, leading actor, director, year, region, language. Among them, the tags under the first-level classification may include: TV series, movies; the tags under the second-level classification may include: comedy, love, drama.

[0071] S102. Determine multiple interested topics of the target user from multiple topics corresponding to multiple topic tags in a topic library according to the multiple topic tags and the preference tags.

[0072] Among them, the topic library may include multiple topic tags, and multiple topics corresponding to each topic tag. Each topic has a corresponding topic id (Identity document, unique code) and a corresponding topic name.

[0073] In some embodiments, multiple theme tags and preference tags in a theme library can be processed to obtain a matching result, and based on the matching result, multiple interested themes of a target user can be determined from multiple themes corresponding to the multiple theme tags.

[0074] It should be noted that the matching of multiple theme tags and preference tags can be performed sequentially, or simultaneously, or in other ways. The embodiments of the present application do not specifically limit this.

[0075] S103. Determine a target theme from multiple interested themes according to the video content in the multiple interested themes.

[0076] Among them, the number of target themes can be at least one.

[0077] In some embodiments, a preset theme screening rule is adopted to screen multiple interested themes according to the video content in the multiple interested themes, and at least one target theme is obtained. Each target theme may include at least one video content.

[0078] In addition, the video content may include multiple consecutive frames of images and audio.

[0079] S104. Recommend the video content in the target theme to the terminal device corresponding to the target user.

[0080] Among them, the server can send the video content in the target theme to the terminal device corresponding to the target user, and the terminal device corresponding to the target user receives the video content in the target theme, and then presents the video content in the target theme to the target user.

[0081] In the embodiments of the present application, the terminal device corresponding to the target user may present a graphical user interface, and the graphical user interface includes a theme recommendation control. The terminal device corresponding to the target user responds to a selection operation on the theme recommendation control input by the target user, generates and sends a theme recommendation request to the server, and the server can execute the processes of S101 to S104 according to the theme recommendation request.

[0082] In addition, the terminal device corresponding to the target user responds to a sliding operation input by the target user to slide and display the graphical user interface. If it slides to the theme recommendation area in the graphical user interface, it generates and sends a theme recommendation request to the server, and the server can execute the processes of S101 to S104 according to the theme recommendation request. Of course, the terminal device corresponding to the target user may also generate a theme recommendation request in other ways, and the embodiments of the present application do not specifically limit this.

[0083] In summary, the embodiment of the present invention provides an information recommendation method, including: determining a preference label of a target user from multiple labels of a historical video according to the operation behavior of the target user on the historical video, where the historical video is a video that the target user has watched historically; determining multiple interested topics of the target user from multiple topics corresponding to multiple topic labels in a topic library according to the multiple topic labels and the preference label; determining a target topic from the multiple interested topics according to the video content in the multiple interested topics; and recommending the video content in the target topic to the terminal device corresponding to the target user. First, the preference label of the target user and the multiple interested topics of the target user are automatically determined, and then the target topic is determined from the multiple interested topics according to the video content in the multiple interested topics, which can automatically determine the target topic without manual screening, saving human resources and improving the efficiency of topic recommendation.

[0084] Optionally, Figure 2 As shown in the flowchart of an information recommendation method provided by an embodiment of the present invention, Figure 2 as shown, the process of determining the preference label of the target user from multiple labels of the historical video according to the operation behavior of the target user on the historical video in S101 above may include:

[0085] S201. Determine the preference scores of the target user for multiple labels in each label dimension according to the operation behavior of the target user on the historical video.

[0086] Among them, the number of label dimensions may be multiple.

[0087] Optionally, the preference scores of multiple labels in multiple label dimensions may include: the preference scores of multiple labels in the first-level classification, the preference scores of multiple labels in the second-level classification, the preference scores of multiple labels of the leading actors, the preference scores of multiple labels of the directors, the preference scores of multiple labels of the eras, the preference scores of multiple labels of the regions, and the preference scores of multiple labels of the languages.

[0088] In addition, the server may calculate the preference scores of the target user for multiple labels in sequence, or calculate the preference scores of the target user for multiple labels simultaneously, or calculate the preference scores of the target user for multiple labels in other ways, and the embodiments of the present application do not specifically limit this.

[0089] S202. Sort the multiple labels in each label dimension according to the preference scores of the multiple labels in each label dimension to obtain the sorted multiple labels in each label dimension.

[0090] In some embodiments, the multiple tags in each tag dimension can be sorted in descending order according to the preference scores of the multiple tags to obtain the sorted multiple tags; alternatively, the multiple tags in each tag dimension can be sorted in ascending order according to the preference scores of the multiple tags to obtain the sorted multiple tags.

[0091] S203. Determine the preferred tag for each tag dimension from the sorted multiple tags in each tag dimension.

[0092] In the embodiments of the present application, the first k tags with the highest preference scores among the sorted multiple tags in each tag dimension are used as the preferred tags for each tag dimension. Here, the first quantity is represented as k.

[0093] Optionally, Figure 3 is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention. As Figure 3 shown, the process of determining the preference scores of the target user for the multiple tags in each tag dimension according to the operation behavior of the target user for the historical videos in S201 above may include:

[0094] S301. For each historical time period, according to the behavior type of the operation behavior, the behavior weight of the behavior type, the number of times the target user performs the operation behavior for each tag among the multiple tags in each tag dimension, the number of times the target user performs the operation behavior for all tags under the target tag dimension corresponding to each tag, the number of times all users perform the operation behavior for all tags under the target tag dimension, and the number of times all users perform the operation behavior for each tag, obtain the initial score of the target user for each tag in each historical time period.

[0095] Among them, the behavior type may include a search behavior type, a play behavior type, and a click behavior type. The multiple historical time periods may be multiple historical time segments, and each historical time period may be represented as T. The operation behaviors in different cycle dimensions are statistically analyzed. Considering the comprehensiveness of theme recommendation, this solution combines the long-term and short-term behavior data of users for recommendation. For example, the multiple historical time segments may include: the last 1 day, the last 7 days, the last 15 days, the last 30 days, etc.

[0096] S302. Accumulate the initial scores for each historical time period to obtain the preference scores of the target user for each tag.

[0097] Among them, a preset first formula can be used to calculate the preference scores of the target user for each tag.

[0098] In some embodiments, the first formula may be:

[0099]

[0100] It should be noted that Score uj represents the preference score of the target user u for the tag j, A represents the set of user behavior types, and w a represents the behavior weight of the behavior with type a, and cnt tuja represents taking the behavior data in the historical time period t, and the number of times the target user u has the behavior of type a for the tag j. all_cnt tua represents taking the behavior data in the historical time period t, and the number of times all tags under the target tag dimension corresponding to the tag j of the target user u have the behavior of type a. all_cnt ta represents taking the operation behavior in the historical time period t, and the number of times all tags under the target tag dimension corresponding to the tag j of all users have the behavior of type a. cnt tja represents taking the operation behavior in the historical time period t, and the number of times all users have the behavior of type a for the tag j.

[0101] It is worth noting that w a represents the behavior weight of the behavior with type a. Different behavior types have different degrees of reflecting user preferences. For example, search and play behaviors are more important than clicks. Therefore, they should also be given higher weights, and here it needs to be configured according to the actual business situation and business experience.

[0102] In addition, the relationship between the behavior weight and the final recommendation effect: Since the behavior weight will participate in the modeling score calculation of user interests, if a user has performed a behavior with a higher weight on video A, the algorithm will assign a higher recommendation weight to other videos A' similar to A, and thus is more inclined to recommend video A'. Therefore, the target behaviors expected by users in the business (such as play > search > click) can be set with higher weights to obtain better recommendation effects.

[0103] Optionally, Figure 4 is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention. As Figure 4 shown, the process of determining multiple interesting topics of the target user from multiple topics corresponding to multiple topic tags according to the multiple topic tags and preference tags in S102 above may include:

[0104] S401. Match the multiple topic tags and preference tags to obtain the interest scores of the target user for multiple topics.

[0105] In some embodiments, a preset second formula may be used to match the multiple topic tags and preference tags to obtain the interest scores of the target user for multiple topics.

[0106] S402. Sort multiple topics according to the interest scores of the multiple topics to obtain the sorted multiple topics.

[0107] In the embodiment of the present application, the multiple topics are sorted in descending order according to the interest scores of the multiple topics to obtain the sorted multiple topics, or the multiple topics are sorted in ascending order according to the interest scores of the multiple topics to obtain the sorted multiple topics.

[0108] S403. Determine multiple interested topics of the target user from the sorted multiple topics.

[0109] Among them, the multiple interested topics form a topic list, that is, topic recall is performed.

[0110] It should be noted that, from the sorted multiple topics, select the second largest number of topics with the highest interest scores as the multiple interested topics of the target user. Among them, the second largest number can be denoted as N. For example, N can be 100.

[0111] Optionally, the process of matching the multiple topic tags and preference tags in S401 to obtain the interest scores of the target user for the multiple topics may include:

[0112] Calculate according to the preset weight corresponding to the tag dimension, the number of matches between the preference tag and each topic tag in the corresponding tag dimension, and the preset value in the tag dimension to obtain the interest scores of the multiple topics.

[0113] In some embodiments, the preset second formula may be:

[0114]

[0115] Among them, Score us represents the interest score of the target user u for the topic s. L represents the set of tag dimensions, which may include first-level classification, second-level classification, leading actor, director, era, region, language, etc. w l represents the preset weight corresponding to the tag dimension l, which can also be called the tag matching weight; according to the actual business situation and business experience, different tag matching weights are configured for the tags in different tag dimensions. cnt l represents the number of matches between the preference tag of the target user u and the topic tag of the topic s in the corresponding tag dimension l. k l represents the preset value of the tag dimension l, and the preset value can be the first number, that is, the number of preference tags when the tag dimension is l.

[0116] Exemplarily, the label matching weight of the first-level classification is 100, and the label matching weight of the second-level classification is 80. The first-level classification label k is 2, and the second-level classification k is 3. The first-level classification preference of the target user u is [TV series, movies], and the second-level classification preference is [love, action, comedy]. The theme s label is: the first-level classification label is movies, and the second-level classification label is [comedy | love]. Then the matching score between the user's preference label and the theme information label is 100 * 1 / 2 + 80 * 2 / 3 = 103.33 points.

[0117] Optionally, Figure 5 is a schematic flowchart of an information recommendation method provided by an embodiment of the present invention. As Figure 5 shown, the process of determining the target theme from multiple interesting themes according to the video content in multiple interesting themes in S103 above may include:

[0118] S501. Score the video content in each interesting theme to obtain the score of each video content in each interesting theme.

[0119] In the embodiments of the present application, using a preset scoring rule, each interesting theme has multiple video contents, and each video content in each interesting theme can be scored to obtain the score of each video content in each interesting theme.

[0120] S502. Determine the target video content of each interesting theme according to the score of each video content in each interesting theme.

[0121] Among them, the target video content of each interesting theme can be called the program content pool of each interesting theme.

[0122] In some embodiments, according to the score of each video content in each interesting theme, the multiple video contents in each interesting theme are sorted in descending order, and the third largest number of video contents with the highest scores in each interesting theme are selected as the target video content of each interesting theme, where the third largest number can be represented as m.

[0123] S503. Obtain the score of each interesting theme according to the score of the target video content of each interesting theme.

[0124] S504. Determine the target theme from multiple interesting themes according to the score of each interesting theme.

[0125] In some embodiments, the average value of the scores of multiple target video contents for each topic of interest is calculated to obtain the score of each topic of interest; according to the scores of each topic of interest, the multiple topics of interest are sorted in descending order, and the fourth largest number of topics of interest with the highest scores among the multiple topics of interest is used as the target topic. Among them, the third quantity can be expressed as n.

[0126] Exemplarily, n can be 100 and m can be 10.

[0127] Optionally, Figure 6 The flowchart of an information recommendation method provided by an embodiment of the present invention is shown in Figure 6 As shown, the process of scoring the video contents in each topic of interest in S501 to obtain the scores of each video content in each topic of interest may include:

[0128] S601. Obtain the feature information of the target user and the feature information of the video contents in each topic of interest.

[0129] Among them, the feature extraction is respectively performed on the target user and the video contents in each topic of interest to obtain the feature information of the target user and the feature information of the video contents in each topic of interest.

[0130] In the embodiments of the present application, the feature information of the target user may include: the basic attributes of the target user (age, gender, region, device information, whether a member, etc.), the behavior sequence (viewing sequence), the interest preference, that is, the preference tags in each label dimension (the first-level classification preference, the second-level classification preference, the leading actor preference, the director preference, the language preference, the region preference, etc.) of the video, the activity and stickiness (the duration from the last visit to the present, the viewing times, the frequency).

[0131] In addition, the feature information of the video contents in each topic of interest may include: the basic attributes and the popularity. Among them, the basic attributes may include: video id, the first and second-level classifications, the leading actor, the director, the language, and the popularity may include: the number of plays, the play uv (texture mapping coordinates), the click-through rate, the play conversion rate, the average play duration per play, and the average play duration per capita.

[0132] S602. Use a preset video content scoring model to process the feature information of the target user and the feature information of the video contents in each topic of interest to obtain the scores of each video content in each topic of interest.

[0133] Among them, the scores of each video content in each topic of interest can be used to characterize the on-demand rate of each video content in each preset topic of interest.

[0134] In some embodiments, a preset video content scoring model is adopted. The feature information of the target user and the feature information of the video content in each interesting topic are extracted as vectors, the vectors are concatenated, and the concatenated vectors are input into a multi-layer DNN (Deep Neural Network). The DNN can output the scores of each video content in each interesting topic. Among them, the DNN is the network structure in the preset video content scoring model.

[0135] Optionally, the method may further include:

[0136] Training a model based on the feature information of the sample user, the first sample video content that has been exposed and clicked by the sample user, and the second sample video content that has been exposed but not clicked by the sample user to obtain a preset video content scoring model.

[0137] Among them, the first sample video content is a positive sample, and the second sample video content is a negative sample.

[0138] It should be noted that the first sample video content and the second sample video content are processed to obtain the feature information of the first sample video content and the feature information of the second sample video content. A preset video content scoring model is obtained by training a model based on the feature information of the sample user, the feature information of the first sample video content, and the feature information of the second sample video content.

[0139] In addition, the preset video content scoring model may include: an input layer, an embedding layer, an mlp (perceptron) layer, and an output layer connected in sequence.

[0140] In the embodiments of the present application, the feature information of the sample user: the basic attributes of the sample user (age, gender, region, device information, whether a member, etc.), the behavior sequence (viewing sequence), interest preferences, that is, the preference labels in each label dimension (preference for the first-level classification of videos, second-level classification preference, leading actor preference, director preference, language preference, region preference, etc.), activity and stickiness (the duration since the last visit, the number of viewing times, frequency).

[0141] The feature information of the first sample video content and the feature information of the second sample video content may include: basic attributes and popularity. Among them, the basic attributes may include: video id, first and second-level classifications, leading actor, director, language, and the popularity may include: number of plays, play uv (texture mapping coordinates), click-through rate, play conversion rate, average play duration per play, average play duration per person.

[0142] It should be noted that if the target user is a new user without search, play, or click behavior, popular themes and popular video content in the program content pool of the theme are taken as the default recommended themes and theme content and presented to the end user. If the target user is an old user with search, play, or click behavior, personalized recommendation results are based on the video content of the target subject in the above S104 and presented to the end user. For the case where the number of recommended items does not meet the requirement of the number of recommended items, the default recommended themes and theme content are taken to make up and then presented to the corresponding terminal device of the target user.

[0143] In summary, by collecting the behavior data of the target user, an interest portrait of the target user for video content, tags of video content, and themes is depicted. Through the matching of theme tags and preference tags of the target user, a fixed number of theme lists are recalled. Through the theme tags, the program content pool of the theme is filtered. Based on the deep learning model, the interest score of the target user for video content is predicted, and a fixed number of theme content that the target user is more interested in is obtained for each theme. By aggregating the interest scores of the target user for the theme content, a theme list that the target user is more interested in is finally obtained. Thus, interesting themes and theme content are recommended to the target user in real time, achieving the theme personalization recommendation effect of one person, one face, improving the experience of the target user, enhancing the recommendation efficiency, and reducing the operation burden.

[0144] The following describes an information recommendation device, a server, a storage medium, etc. for executing the information recommendation method provided in this application. For the specific implementation process and technical effects, refer to the relevant content of the above information recommendation method, which will not be elaborated below.

[0145] Figure 7 The following is a schematic structural diagram of an information recommendation device provided by an embodiment of the present invention, as Figure 7 shown. The device may include:

[0146] A determination module 701, configured to determine preference tags of the target user from multiple tags of the historical video according to the operation behavior of the target user for the historical video, where the historical video is a video that the target user has watched historically; determine multiple interested themes of the target user from multiple themes corresponding to the multiple theme tags according to the multiple theme tags in the theme library and the preference tags; and determine a target theme from the multiple interested themes according to the video content in the multiple interested themes;

[0147] A recommendation module 702, configured to recommend video content in the target theme to a terminal device corresponding to the target user.

[0148] Optionally, the determining module 701 is specifically configured to determine, according to the operation behavior of the target user for the historical video, the preference scores of the target user for multiple tags in each tag dimension; sort the multiple tags in each tag dimension according to the preference scores of the multiple tags in each tag dimension to obtain the sorted multiple tags in each tag dimension; and determine the preferred tags in each tag dimension from the sorted multiple tags in each tag dimension.

[0149] Optionally, for each historical time period, the determining module 701 is specifically configured to obtain the initial score of the target user for each tag according to the behavior type of the operation behavior, the behavior weight of the behavior type, the number of times the target user performs the operation behavior for each tag in the multiple tags in each tag dimension, the number of times the target user performs the operation behavior for all tags in the target tag dimension corresponding to each tag, the number of times all users perform the operation behavior for all tags in the target tag dimension, and the number of times all users perform the operation behavior for each tag; and accumulate the initial scores for each historical time period to obtain the preference scores of the target user for the multiple tags in each tag dimension.

[0150] Optionally, the determining module 701 is specifically configured to match the multiple theme tags and the preferred tags to obtain the interest scores of the target user for the multiple themes; sort the multiple themes according to the interest scores of the multiple themes to obtain the sorted multiple themes; and determine the multiple interested themes of the target user from the sorted multiple themes.

[0151] Optionally, the determining module 701 is specifically configured to calculate according to the preset weight corresponding to the tag dimension, the matching quantity between the preferred tag and each theme tag in the corresponding tag dimension, and the preset value in the tag dimension to obtain the interest scores of the multiple themes.

[0152] Optionally, the determining module 701 is specifically configured to score the video content in each interested theme to obtain the scores of each video content in each interested theme; determine the target video content of each interested theme according to the scores of each video content in each interested theme; obtain the scores of each interested theme according to the scores of the target video content of each interested theme; and determine the target theme from the multiple interested themes according to the scores of each interested theme.

[0153] Optionally, the determining module 701 is specifically configured to obtain the feature information of the target user and the feature information of the video content in each of the interested topics; use a preset video content scoring model to process the feature information of the target user and the feature information of the video content in each of the interested topics, and obtain the score of each video content in each of the interested topics.

[0154] Optionally, the apparatus further includes:

[0155] A training module, configured to perform model training according to the feature information of the sample user, the first sample video content that has been exposed and played by the sample user, and the second sample video content that has been exposed but not played by the sample user, to obtain the preset video content scoring model.

[0156] The above apparatus is used to execute the method provided in the foregoing embodiment, and its implementation principle and technical effects are similar, and will not be described in detail here.

[0157] The above modules may be one or more integrated circuits configured to implement the above method, for example: one or more application specific integrated circuits (ASICs), or, one or more digital signal processors (DSPs), or, one or more field programmable gate arrays (FPGAs), etc. Again, when the above certain module is implemented in the form of a processing element dispatching program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processors that can call program code. Again, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0158] Figure 8 A schematic structural diagram of a server provided by an embodiment of the present invention is shown as Figure 8 shown, the server may include: a processor 801, a memory 802.

[0159] The memory 802 is used to store a program, and the processor 801 calls the program stored in the memory 802 to execute the above method embodiment. The specific implementation manner and technical effects are similar, and will not be described in detail here.

[0160] Optionally, the present invention further provides a program product, such as a computer-readable storage medium, including a program, which is used to execute the above method embodiment when being executed by a processor.

[0161] In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0162] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0163] In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of a combination of hardware and software functional units.

[0164] The above-mentioned integrated units implemented in the form of software functional units can be stored in a computer-readable storage medium. The above-mentioned software functional units stored in a storage medium include several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor (English: processor) to execute some steps of the methods described in each embodiment of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (English: Read-Only Memory, abbreviated as: ROM), random access memories (English: Random Access Memory, abbreviated as: RAM), magnetic disks or optical discs that can store program codes.

[0165] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An information recommendation method, characterized in that, Including: For each historical time period, based on the operation behaviors of the target user for historical videos, obtain the initial scores of the target user for each tag in each historical time period; Accumulate the initial scores of each historical time period, and use a preset first formula to calculate the preference scores of the target user for multiple tags in each tag dimension; wherein, the preset first formula is: Among them, the represents the preference score of the target user u for the label j, the A represents the set of user behavior types, and the represents the behavior weight of the behavior type a, and the represents the number of times the target user u has the behavior type a for the label j in the historical time period t; the represents the number of times the target user u has the behavior type a for all labels under the target label dimension corresponding to the label j in the historical time period t; the represents the number of times all users have the behavior type a for all labels under the target label dimension corresponding to the label j in the historical time period t; the represents the number of times all users have the behavior type a for the label j in the historical time period t; Sort the multiple tags in each tag dimension according to the preference scores of the multiple tags in each tag dimension, and obtain the sorted multiple tags in each tag dimension; from the sorted multiple tags in each tag dimension, determine the preferred tags in each tag dimension; According to the matching quantity between the preferred tags in the theme library and each theme tag in the corresponding tag dimension, the preset value in the tag dimension, and the preset weight corresponding to the tag dimension, calculate using a preset second formula to obtain the interest scores of the multiple themes; wherein, the preset second formula is: Among them, the represents the interest score of the target user u in the theme s; the L represents the set of label dimensions; the represents the corresponding preset weight when the label dimension is ; the represents the number of matches between the preference label of the target user u and the theme label of the theme s in the corresponding label dimension ; the represents the number of preference labels when the label dimension is . Sort the multiple themes according to the interest scores of the multiple themes, and obtain the sorted multiple themes; from the sorted multiple themes, determine the multiple interested themes of the target user; According to the video content in the multiple interested themes, determine the target theme from the multiple interested themes; recommend the video content in the target theme to the terminal device corresponding to the target user.

2. The method according to claim 1, wherein The obtaining of the initial scores of the target user for each tag in each historical time period based on the operation behaviors of the target user for historical videos includes: Based on the behavior type of the operation behavior, the behavior weight of the behavior type, the number of times the target user performs the operation behavior for each tag in the multiple tags in each tag dimension of the target user, the number of times the target user performs the operation behavior for all tags in the target tag dimension corresponding to each tag, the number of times all users perform the operation behavior for all tags in the target tag dimension, and the number of times all users perform the operation behavior for each tag, obtain the initial scores of the target user for each tag in each historical time period; wherein, the behavior type includes: search behavior type, play behavior type, click behavior type.

3. The method according to claim 1, wherein The determining of the target theme from the multiple interested themes according to the video content in the multiple interested themes includes: Rate the video content in each interested theme to obtain the scores of each video content in each interested theme; Based on the scores of each video content in each interested theme, determine the target video content of each interested theme; Based on the scores of the target video content of each interested theme, obtain the scores of each interested theme; Based on the scores of each interested theme, determine the target theme from the multiple interested themes.

4. The method according to claim 3, characterized in that, The rating of the video content in each interested theme to obtain the scores of each video content in each interested theme includes: Obtain the feature information of the target user and the feature information of the video content in each of the interested topics; Use a preset video content scoring model to process the feature information of the target user and the feature information of the video content in each of the interested topics, and obtain the scores of each video content in each of the interested topics.

5. The method according to claim 4, wherein The method further includes: Perform model training based on the feature information of the sample user, the first sample video content that has been exposed and played by the sample user, and the second sample video content that has been exposed but not played by the sample user to obtain the preset video content scoring model.

6. An information recommendation device, characterized in that, Includes: A determination module, configured to, for each historical time period, obtain the initial score of the target user for each tag according to the operation behavior of the target user for the historical videos in each historical time period; Accumulate the initial scores for each historical time period, and use a preset first formula to calculate the preference scores of the target user for multiple tags in each tag dimension; wherein, the preset first formula is: Among them, the represents the preference score of the target user u for the label j, the A represents the set of user behavior types, and the represents the behavior weight of the behavior type a, and the represents the number of times the target user u has the behavior type a for the label j at the historical time period t; the represents the number of times the target user u has the behavior type a for all labels under the target label dimension corresponding to the label j at the historical time period t; the represents the number of times all users have the behavior type a for all labels under the target label dimension corresponding to the label j at the historical time period t; the represents the number of times all users have the behavior type a for the label j at the historical time period t; Sort the multiple tags in each tag dimension according to the preference scores of the multiple tags in each tag dimension to obtain the sorted multiple tags in each tag dimension; determine the preferred tags in each tag dimension from the sorted multiple tags in each tag dimension; According to the matching quantity between the preferred tags in the theme library and each theme tag in the corresponding tag dimension, the preset value in the tag dimension, and the preset weight corresponding to the tag dimension, use a preset second formula to calculate and obtain the interest scores of the multiple themes; wherein, the preset second formula is: Among them, the represents the interest score of the target user u in the theme s; the L represents the set of label dimensions; the represents the corresponding preset weight when the label dimension is ; the represents the number of matches between the preference label of the target user u and the theme label of the theme s on the corresponding label dimension ; the represents the number of the preference labels when the label dimension is . Sort the multiple themes according to the interest scores of the multiple themes to obtain the sorted multiple themes; determine the multiple interested themes of the target user from the sorted multiple themes; determine the target theme from the video content in the multiple interested themes; A recommendation module, configured to recommend the video content in the target theme to the terminal device corresponding to the target user.

7. A server, characterized in that, Includes: A memory and a processor, the memory stores a computer program executable by the processor, and when the processor executes the computer program, it implements the information recommendation method according to any one of claims 1-5 above.

Images