WO2022068492A1 - 视频推荐方法及装置 - Google Patents

视频推荐方法及装置 Download PDF

Info

Publication number
WO2022068492A1
WO2022068492A1 PCT/CN2021/115027 CN2021115027W WO2022068492A1 WO 2022068492 A1 WO2022068492 A1 WO 2022068492A1 CN 2021115027 W CN2021115027 W CN 2021115027W WO 2022068492 A1 WO2022068492 A1 WO 2022068492A1
Authority
WO
WIPO (PCT)
Prior art keywords
video
tag
weight
candidate
videos
Prior art date
Application number
PCT/CN2021/115027
Other languages
English (en)
French (fr)
Inventor
赵会铸
Original Assignee
百果园技术(新加坡)有限公司
赵会铸
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 百果园技术(新加坡)有限公司, 赵会铸 filed Critical 百果园技术(新加坡)有限公司
Priority to EP21874153.6A priority Critical patent/EP4224874A4/en
Priority to US18/246,004 priority patent/US20230362423A1/en
Publication of WO2022068492A1 publication Critical patent/WO2022068492A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/251—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4668—Learning process for intelligent management, e.g. learning user preferences for recommending movies for recommending content, e.g. movies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/25—Management operations performed by the server for facilitating the content distribution or administrating data related to end-users or client devices, e.g. end-user or client device authentication, learning user preferences for recommending movies
    • H04N21/258—Client or end-user data management, e.g. managing client capabilities, user preferences or demographics, processing of multiple end-users preferences to derive collaborative data
    • H04N21/25866—Management of end-user data
    • H04N21/25891—Management of end-user data being end-user preferences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/4508—Management of client data or end-user data
    • H04N21/4532—Management of client data or end-user data involving end-user characteristics, e.g. viewer profile, preferences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4667—Processing of monitored end-user data, e.g. trend analysis based on the log file of viewer selections

Definitions

  • the present application relates to the technical field of automatic recommendation, for example, to a video recommendation method and device.
  • Short video applications can provide users with rich audio and video content, and the underlying recommendation system undertakes the important work of screening audio and video that users are interested in from the massive resources of the platform.
  • the recommender system needs to be able to capture the user's interest points in real time, and select short video content that matches it from the candidate set, so as to improve the user experience.
  • the recommendation system captures user interest points in real time.
  • the common solution is to introduce user session data in the model training process in the recall and sorting stages.
  • the problem of candidate sets but due to the delay in model update, the control over the final video delivery is not enough, and there is often insufficient or excess delivery of interest points.
  • the present application provides a video recommendation method and device, so as to solve the problem in the related art that the recommendation system has insufficient control over the final delivered video, and the delivery of points of interest is insufficient or excessive.
  • the application provides a video recommendation method, the method includes:
  • the candidate set contains the video tags that meet the set conditions and has not been issued.
  • the video is downgraded.
  • the present application also provides a video recommendation device, the device comprising:
  • the label weight determination module is set to obtain the video mark of the historical operation of the target account, and determines the label weight of each video label in the multiple video labels contained in the video corresponding to the video mark;
  • the video weight determination module is set to determine the video weight of each candidate video in the plurality of candidate videos in the preset candidate set based on the label weight of each video label;
  • a video recommendation module configured to select a target video from the candidate set for recommendation according to the video weight of each candidate video
  • the weight reduction processing module is configured to detect the video tags that meet the set conditions in the videos that have been issued, and according to the tag weight of the video tags that meet the set conditions, the candidate set contains the videos that meet the set conditions. The candidate videos that are tagged and not issued will be downgraded.
  • the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, where the processor implements the above-mentioned video recommendation method when executing the program.
  • the present application also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, implements the above-mentioned video recommendation method.
  • FIG. 1 is a flowchart of an embodiment of a video recommendation method provided in Embodiment 1 of the present application;
  • FIG. 2 is a flowchart of an embodiment of a video recommendation method provided in Embodiment 2 of the present application;
  • FIG. 3 is a structural block diagram of an embodiment of a video recommendation apparatus provided in Embodiment 3 of the present application;
  • FIG. 4 is a schematic structural diagram of an electronic device provided in Embodiment 4 of the present application.
  • FIG. 1 is a flowchart of an embodiment of a video recommendation method provided in Embodiment 1 of the present application.
  • This embodiment can be applied to a video recommendation scenario.
  • this embodiment can be implemented by a video recommendation device.
  • the video recommendation apparatus can be applied in a video application program, a short video application program or a live broadcast application program to perform video recommendation.
  • Step 110 Acquire the video identifier of the historical operation of the target account, and determine the tag weight of each video tag in the multiple video tags included in the video corresponding to the video identifier.
  • the video identification of the previous operation of the target account may be obtained from the log record.
  • a video may include one or more video tags, and a type of video tag may appear in one or more videos, in one embodiment, statistics in the video corresponding to the video ID of the historical operation of the target account may be performed. The number of occurrences of each video tag, and then normalizing the number of occurrences of each video tag to obtain the tag weight of each video tag.
  • Step 120 Determine the video weight of each candidate video in the plurality of candidate videos in the preset candidate set based on the label weight of each video label.
  • the candidate set may be a set composed of one or more candidate videos that are determined according to a general recommendation method and match the target account.
  • the video of the candidate video related to the historical video tag in the candidate set can be determined according to the tag weight of each video tag appearing in the history Weights.
  • the video tag included in the candidate video can be matched with the video tag whose tag weight has been determined, so as to determine the video tag included in the candidate video.
  • label weight For a video tag with no matching item in the candidate video, the weight of the video tag can be set to a preset value, for example, set to a value of 0.
  • the video weight of the candidate video can be calculated according to the tag weight of the video tags contained in the candidate video, for example, the sum of the tag weights of the video tags contained in the candidate video is calculated as the video weight.
  • Step 130 Select a target video from the candidate set for recommendation according to the video weight of each candidate video.
  • At least one candidate video with the largest video weight may be selected as the target video, and the target video recommendation can be delivered to the client.
  • Step 140 Detect the video tags that meet the set conditions in the videos that have been issued, and according to the tag weight of the video tags that meet the set conditions, for the video tags that meet the set conditions in the candidate set, those that have not been issued
  • the candidate videos are down-weighted.
  • the number of delivered videos corresponding to each video tag can be monitored, and if a video tag satisfies the set conditions, the subsequent video delivery volume of the video tag can be limited to avoid full screen The case of a single point of interest video.
  • the video tags whose video delivery volume meets the set conditions and the corresponding un-delivered videos can be processed for weight reduction.
  • the un-delivered videos can be quickly withdrawn from delivery. , so as to realize the dynamic recommendation strategy for the delivered video.
  • the video weight of each candidate video in the candidate set is determined by obtaining the tag weight of each video tag included in the video corresponding to the video identifier of the historical operation of the target account.
  • the target video After the target video is recommended, it can detect the video tags that meet the set conditions in the videos that have been delivered.
  • the candidate set is selected.
  • the unpublished video that contains the video tag that meets the set conditions is processed for weight reduction, so as to realize the dynamic adjustment of the video weight, ensure the timely and effective distribution of the user's real-time point of interest video, and avoid the full screen of a single interest. Clicking on the video can improve the user's experience of using video applications, including increasing the usage time, usage frequency and other indicators.
  • FIG. 2 is a flowchart of an embodiment of a video recommendation method provided in Embodiment 2 of the present application. This embodiment is described on the basis of Embodiment 1, and includes the following steps:
  • Step 210 Acquire historical behavior data of the target account, where the historical behavior data includes video identification and operation information of the operation of the target account.
  • this embodiment may further include the following steps:
  • a video pull request sent by the client is received, where the video pull request includes the target account.
  • the client can detect the user's operation of pulling the video, and after detecting the operation, send a video pulling request to the video recommendation device, and the video pulling request can include the target account.
  • the video recommendation device parses the request to obtain a target account, and obtains historical behavior data of the target account.
  • the historical behavior data may be the behavior data of the last session (Session) of the target account, and the real-time performance of the historical behavior data of the last session can reach the second level, which can improve the real-time performance of real-time interest video recommendation.
  • the last session (Session) refers to the time interval during which the target account communicates with the server of the application program for the last time (ie, the last time).
  • a storage area may be allocated for each session for storing the data of the session operation.
  • the behavior data of each session of the user can be stored in the storage area of the current session, so the historical behavior data of the target user can be obtained from the storage area of the most recent session of the target account.
  • V ⁇ v 1 ,v 2 ,...,v k ⁇ .
  • the operation information may include positive operation information and negative operation information, wherein the positive operation information may include like, finish broadcasting, sharing, sending gifts, etc., and the negative operation information may include stepping on, thumbs down, and the like.
  • Step 220 Acquire video tags included in videos corresponding to multiple video identifiers, and determine distribution information of each video tag according to the operation information.
  • a video may include one or more video tags (Tag and hashtag, etc.), and the video tag corresponding to each video tag in the video sequence can be obtained, and the video tags of each video can be formed into a video tag sequence.
  • the distribution information of each video tag can be determined in combination with the operation information of each video.
  • the step of determining the distribution information of each video tag according to the operation information in step 220 may include the following steps:
  • For each video tag count the first number of times that the video tag appears positive operation information and the second number of times that negative operation information appears, as the distribution information.
  • a video tag may have both positive action information and negative action information.
  • the user likes video v 1 and downvotes video v 2 , and both videos have tags t 1 .
  • the user has liked the videos v 1 and v.
  • the video v 3 is downvoted, the video v 1 includes tags t 1 , t 2 and t 3 , the video v 2 includes t 2 and t 4 , and the video v 3 includes t 2 and t 3 , then for the tag t 1 ,
  • the first number of times that the forward operation information appears is 1, and the second number of times that the negative operation information appears is 0; for the label t 2 , the first number of times that the forward operation information appears is 2, and the negative operation information appears.
  • the second number of times is 1; for label t 3 , the first number of times that positive operation information appears is 1, and the second number of times that negative operation information appears is 1; for label t 4 , the occurrence of positive operation information is 1.
  • the first occurrence of the message is 1, and the second occurrence of the negative operation message is 0.
  • the first number of times the positive operation information appears in the video tag and the second number of times the negative operation information appears in the video tag can be used as the distribution information of the video tag.
  • Step 230 Determine the tag weight of the corresponding video tag according to the distribution information.
  • step 230 may include the following steps:
  • Step 230-1 Calculate the weight reference value of the corresponding video tag according to the first number of times and the second number of times.
  • the weight reference value of the corresponding video tag can be calculated in the following manner:
  • Weight reference value (k1+1)/(k1+k2+1); wherein, k1 is the first order, and k2 is the second order.
  • Step 230-2 if the weight reference value is greater than the first preset value, determine the video tag as a forward video tag, and determine the tag weight of the video tag as the weight reference value.
  • Step 230-3 if the weight reference value is less than the second preset value, then determine the video tag as a negative video tag, and determine the tag weight of the video tag as the negative number of the weight reference value, wherein , the second preset value is smaller than the first preset value.
  • this embodiment may further include: adding the forward video tag to a set of forward video tags.
  • this embodiment may further include: adding the negative video tag to a set of negative video tags.
  • the video label After the weight reference value of the video label is obtained, if the weight reference value is greater than the first preset value, the video label can be determined as a forward video label, and the video label is added to the forward video label set.
  • the tag weight of the video tag is determined as the weight reference value. If the weight reference value is less than the second preset value, the video tag may be determined as a negative video tag, and the video tag may be added to the negative video tag set, and at the same time, the tag weight of the video tag is determined as the above-mentioned video tag. Negative number for the weight reference value. Wherein, the second preset value is smaller than the first preset value.
  • video tags are divided into positive video tag sets and negative video tag sets.
  • the video tags in the positive video tag set tend to be positive behaviors, such as like, finish, share, etc.; negative video tags
  • the video tags in the collection are biased towards negative behaviors, such as stepping, thumbing down, etc.
  • This division method can facilitate the subsequent determination of whether the video tag's weight-escalation score is positive or negative, so that the right-raising or lowering is decided during the video recommendation process. rights handling policy.
  • Step 240 for each video tag, multiply the tag weight of the video tag by a preset weight escalation coefficient to obtain the weight escalation score of the video tag.
  • a weight escalation coefficient can be set according to experience. For example, the weight escalation coefficient is set to a constant between 1 and 5, and the tag weight of each video tag is multiplied by the preset weight escalation coefficient. Get the escalation score for this video tag. Among them, for the video tag in the positive video tag set, since its tag weight is positive, the corresponding weighting score is also positive; for the video tag in the negative video tag set, since its tag weight is negative, then The corresponding escalation score is also a negative number
  • Step 250 for each candidate video in the candidate set, calculate the sum of the weight promotion scores of the video tags included in the candidate video to obtain the video weight of the candidate video.
  • the candidate set is a set composed of multiple candidate videos preliminarily screened by the recommendation device for the target account according to a general method.
  • the content of the candidate set is often broad, that is, it contains more candidate videos that users are interested in, but due to limited pit resources, it is necessary to screen out the content with the best real-time performance.
  • the video tag carried by the candidate video can be obtained, and then each video tag of the candidate video can be traversed to determine whether each video tag has a corresponding weighting score. If the video tag has a weighting score, The right-escalation score will be obtained; if a video tag does not have a right-escalation score, the right-escalation score of the video label can be set to a value of 0. Then calculate the sum of the weighting scores of all the video tags of the candidate video to obtain the video weight of the candidate video.
  • Step 260 according to the video weight of each candidate video, select a target video from the candidate set for recommendation.
  • multiple candidate videos in the candidate set can be sorted according to the video weight, and then one or more candidate videos with the highest video weight are selected as the target video, And send the target video recommendation to the client.
  • Step 270 Detect the video tags that meet the set conditions in the videos that have been issued, and according to the tag weight of the video tags that meet the set conditions, the candidate set contains the video tags that meet the set conditions and has not been issued.
  • the candidate videos are down-weighted.
  • the number of delivered videos corresponding to each video tag can be counted, and when the number of delivered videos corresponding to a video tag reaches a preset number threshold, it can be determined that the video tag satisfies the set If the conditions are met, the unpublished videos that contain the video tag in the candidate set can be down-weighted to limit the subsequent video distribution of the video tag and avoid the situation where the full screen is full of videos of a single point of interest.
  • step 270 may include the following steps:
  • the video weight of the undistributed video that contains the video tag that meets the set condition is subtracted from the weight of the video tag that meets the set condition in the candidate set, to obtain a new video of the candidate video. Weights.
  • the implementation process of weight reduction may be as follows: for the candidate videos that have not been distributed in the candidate set, it can be determined whether the undistributed candidate videos have video tags that meet the set conditions, and if the undistributed candidate videos have video tags that meet the set conditions, If the video has a video tag that satisfies the set condition, then the candidate video with the video tag that satisfies the set condition is subjected to weight reduction processing, that is, the video weight of the candidate video is subtracted from the video weight of the video tag that satisfies the set condition. The new video weight of the candidate video, so as to realize the dynamic adjustment of the video weight of the candidate video.
  • the weight of all videos with video tag A in the candidate set (herein referred to as A-type videos) can be increased by 1 point
  • the weight of all videos with video tag B in the candidate set (herein referred to as B-type videos)
  • all videos with video tag C in the candidate set (herein referred to as C-type videos) are weighted by 0.1 points.
  • the weighting of the remaining A-type videos in the candidate set will be cancelled, that is, the remaining A-type videos will be reduced by 1 point;
  • the B-type videos are delivered to meet the demand , then cancel the weighting of the remaining B-type videos in the candidate set, that is, reduce the weight of the remaining B-type videos by 0.5 points;
  • the C-type videos are delivered to meet the requirements, cancel the weighting of the remaining C-type videos in the candidate set.
  • the remaining category C videos are downgraded by 0.1 points.
  • this embodiment determines the tag weight of each video tag according to the video tag carried by the video corresponding to the video tag in the short-term historical behavior data of the target account, and then determines the video according to the tag weight.
  • the weight of each candidate video in the candidate set can be calculated based on the weight escalation score of each video tag in real time, so as to ensure that the real-time point of interest videos of users can be delivered in time and improve the probability of video delivery.
  • the unpublished video that contains the video tag in the candidate set can be down-weighted according to the tag weight of the video tag. Dynamically adjust the video weight of each candidate video in the candidate set.
  • FIG. 3 is a structural block diagram of an embodiment of a video recommendation apparatus provided in Embodiment 3 of the present application.
  • the video recommendation apparatus may be located in a server and may include the following modules:
  • the tag weight determination module 310 is configured to obtain the video identification of the historical operation of the target account, and determine the tag weight of each video tag in the multiple video tags included in the video corresponding to the video identification; the video weight determination module 320 is configured to set In order to determine the video weight of each candidate video in the plurality of candidate videos in the preset candidate set based on the label weight of each video label; the video recommendation module 330 is set to be based on the video weight of each candidate video, from the candidate set.
  • the weight reduction processing module 340 is configured to detect the video tags that meet the set conditions in the videos that have been delivered, and according to the tag weights of the video tags that meet the set conditions, the candidate set contains all the The unpublished candidate videos of the video tags that meet the set conditions are downgraded.
  • the above-mentioned video recommendation apparatus provided by the embodiment of the present application can execute the video recommendation method provided by any embodiment of the present application, and has functional modules and effects corresponding to the execution method.
  • FIG. 4 is a schematic structural diagram of an electronic device provided in Embodiment 4 of the present application.
  • the electronic device includes a processor 410, a memory 420, an input device 430, and an output device 440;
  • the number can be one or more.
  • one processor 410 is used as an example; the processor 410, the memory 420, the input device 430 and the output device 440 in the electronic device can be connected through a bus or other means. Take bus connection as an example.
  • the memory 420 may be configured to store software programs, computer-executable programs, and modules, such as program instructions/modules corresponding to the video recommendation method in the embodiments of the present application.
  • the processor 410 executes various functional applications and data processing of the electronic device by running the software programs, instructions and modules stored in the memory 420, that is, to implement the above video recommendation method.
  • Embodiment 5 of the present application further provides a storage medium including computer-executable instructions, where the computer-executable instructions are used to perform the video recommendation in any one of Embodiments 1 to 2 when executed by a processor of a server method.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computing Systems (AREA)
  • Computer Graphics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本文公开了一种视频推荐方法及装置。所述视频推荐方法包括：获取目标账户历史操作的视频标识，并确定所述视频标识对应的视频中包含的多个视频标签中每个视频标签的标签权重；基于每个视频标签的标签权重，确定预设的候选集中多个候选视频中每个候选视频的视频权重；根据每个候选视频的视频权重，从所述候选集中选取目标视频进行推荐；检测已下发的视频中满足设定条件的视频标签，按照所述满足设定条件的视频标签的标签权重，对所述候选集中包含所述满足设定条件的视频标签且未下发的候选视频进行降权处理。

Description

视频推荐方法及装置
本申请要求在2020年09月29日提交中国专利局、申请号为202011052701.1的中国专利申请的优先权，该申请的全部内容通过引用结合在本申请中。
技术领域
本申请涉及自动化推荐技术领域，例如涉及一种视频推荐方法及装置。
背景技术
短视频应用程序能够为用户提供丰富的音视频内容，而底层的推荐系统承担了从平台海量资源中筛选用户感兴趣的音视频的重要工作。为了满足不同用户不断变化的兴趣偏好，推荐系统需要能够实时的捕捉到用户兴趣点，并从候选集中筛选出与之匹配的短视频内容，以提升用户的使用体验。
在相关技术中，推荐系统为实时捕捉用户兴趣点，常用的解决方法是在召回、排序阶段在模型训练过程中引入用户会话(session)数据，这在一定程度上能够解决根据用户实时兴趣点决策候选集的问题，但由于模型更新延迟，导致对最终下发视频的控制力不够，往往会出现兴趣点下发不足或过剩的情况。
发明内容
本申请提供一种视频推荐方法及装置，以解决相关技术中推荐系统对最终下发视频的控制力不够，出现兴趣点下发不足或过剩的情况的问题。
本申请提供了一种视频推荐方法，所述方法包括：
获取目标账户历史操作的视频标识，并确定所述视频标识对应的视频中包含的多个视频标签中每个视频标签的标签权重；
基于每个视频标签的标签权重，确定预设的候选集中多个候选视频中每个候选视频的视频权重；
根据每个候选视频的视频权重，从所述候选集中选取目标视频进行推荐；
检测已下发的视频中满足设定条件的视频标签，按照所述满足设定条件的视频标签的标签权重，对所述候选集中包含所述满足设定条件的视频标签且未下发的候选视频进行降权处理。
本申请还提供了一种视频推荐装置，所述装置包括：
标签权重确定模块，设置为获取目标账户历史操作的视频标识，并确定所 述视频标识对应的视频中包含的多个视频标签中每个视频标签的标签权重；
视频权重确定模块，设置为基于每个视频标签的标签权重，确定预设的候选集中多个候选视频中每个候选视频的视频权重；
视频推荐模块，设置为根据每个候选视频的视频权重，从所述候选集中选取目标视频进行推荐；
降权处理模块，设置为检测已下发的视频中满足设定条件的视频标签，按照所述满足设定条件的视频标签的标签权重，对所述候选集中包含所述满足设定条件的视频标签且未下发的候选视频进行降权处理。
本申请还提供了一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现上述的视频推荐方法。
本申请还提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现上述的视频推荐方法。
附图说明
图1是本申请实施例一提供的一种视频推荐方法实施例的流程图；
图2是本申请实施例二提供的一种视频推荐方法实施例的流程图；
图3是本申请实施例三提供的一种视频推荐装置实施例的结构框图；
图4是本申请实施例四提供的一种电子设备的结构示意图。
具体实施方式
下面结合附图和实施例对本申请进行说明。
实施例一
图1是本申请实施例一提供的一种视频推荐方法实施例的流程图，本实施例可以应用于视频推荐场景，例如，应用于短视频推荐的场景中，本实施例可以由视频推荐装置实现，该视频推荐装置可以应用在视频应用程序或短视频应用程序或直播应用程序中，进行视频推荐。
本实施例可以包括如下步骤：
步骤110，获取目标账户历史操作的视频标识，并确定所述视频标识对应的视频中包含的多个视频标签中每个视频标签的标签权重。
在一种实现中，可以从日志记录中获取目标账户在先操作的视频标识。
由于一个视频可以包括一个或多个视频标签，而一种视频标签可以出现在一个或多个视频中，因此，在一种实施方式中，可以统计目标账户历史操作的视频标识对应的视频中的每个视频标签出现的次数，然后对每个视频标签出现的次数进行归一化等处理，从而获得每个视频标签的标签权重。
步骤120，基于每个视频标签的标签权重，确定预设的候选集中多个候选视频中每个候选视频的视频权重。
该步骤中，候选集可以为根据通用的推荐方法确定的、与目标账户匹配的一个或多个候选视频组成的集合。
当通过步骤110确定历史操作记录中出现的每个视频标签的标签权重以后，则可以根据历史出现的每个视频标签的标签权重，来确定候选集中与历史出现的视频标签相关的候选视频的视频权重。
由于一个视频可以包括一个或多个视频标签，则对于候选集中每个候选视频，可以将该候选视频包含的视频标签与已确定标签权重的视频标签进行匹配，从而确定该候选视频包含的视频标签的标签权重。对于候选视频中不存在匹配项的视频标签，可以将该视频标签的权重设置为预设值，例如，设置为数值0。
然后可以根据候选视频包含的视频标签的标签权重来计算该候选视频的视频权重，例如，计算候选视频包含的视频标签的标签权重之和作为视频权重。
步骤130，根据每个候选视频的视频权重，从所述候选集中选取目标视频进行推荐。
在一种实现中，当确定候选集中每个候选视频的视频权重以后，则可以选取视频权重最大的至少一个候选视频作为目标视频，并将目标视频推荐下发至客户端。
步骤140，检测已下发的视频中满足设定条件的视频标签，按照该满足设定条件的视频标签的标签权重，对所述候选集中包含该满足设定条件的视频标签的、未下发的候选视频进行降权处理。
在该实施例中，可以监测每个视频标签对应的已下发视频的数量，如果一视频标签满足设定条件，则可以限制后续对该视频标签的视频下发量，以避免满屏全是单一兴趣点视频的情况。
在实现时，则可以对视频下发量满足设定条件的视频标签、对应的未下发视频进行降权处理，通过降低未下发视频的权重，来使得未下发视频能够快速退出下发，从而实现对下发视频的动态推荐策略。
在本实施例中，通过获取目标账户历史操作的视频标识对应的视频中包含 的每个视频标签的标签权重，来确定候选集中每个候选视频的视频权重，当根据上述视频权重从候选集中选取目标视频进行推荐以后，可以检测已下发的视频中满足设定条件的视频标签，当检测到满足设定条件的视频标签时，按照该满足设定条件的视频标签的标签权重，对候选集中包含该满足设定条件的视频标签的、未下发的视频进行降权处理，从而实现视频权重的动态调整，确保用户实时兴趣点视频的及时有效下发的同时，避免满屏全是单一兴趣点视频的情况，能够提升用户使用视频应用的体验，包括增加使用时长、使用频率等指标。
实施例二
图2是本申请实施例二提供的一种视频推荐方法实施例的流程图，本实施例在实施例一的基础上进行说明，包括如下步骤：
步骤210，获取目标账户的历史行为数据，所述历史行为数据包括所述目标账户操作的视频标识以及操作信息。
在一种可能的场景中，在步骤210之前，本实施例还可以包括如下步骤：
接收客户端发送的视频拉取请求，所述视频拉取请求包括目标账户。
当用户打开视频应用程序并登录目标账户以后，客户端可以检测用户拉取视频的操作，并在检测到该操作后向视频推荐装置发出视频拉取请求，该视频拉取请求可以包括目标账户。视频推荐装置接收到该视频拉取请求后，对该请求进行解析以获得目标账户，并获取目标账户的历史行为数据。
在一种示例中，历史行为数据可以为目标账户最近一次会话(Session)的行为数据，采用最近一次会话的历史行为数据的实时性能达到秒级别，可以提升实时兴趣视频推荐的实时性。其中，最近一次会话(Session)是指目标账户与应用程序的服务器最近一次(即上一次)进行通信的时间间隔。在一种实现中，为了提升用户的访问速度，可以为每次会话分配存储区用于存储该次会话操作的数据。则在视频应用程序中，用户每次会话的行为数据都可以存储在当次会话的存储区中，因此可以从目标账户最近一次会话的存储区中获取目标用户的历史行为数据。
示例性地，历史行为数据可以包括目标账户操作的视频标识以及操作信息，目标账户在最近一次会话中操作的视频标识可以组织成视频序列V＝{v 1,v 2,…,v k}。例如，如果目标用户在上一次会话中共操作过100个视频，则视频序列为V＝{v 1,v 2,…,v 100}。
操作信息可以包括正向操作信息以及负向操作信息，其中，正向操作信息可以包括点赞、完播、分享、送礼物等，负向操作信息可以包括踩、倒拇指等。
步骤220，获取多个视频标识对应的视频所包含的视频标签，并根据所述操作信息确定每个视频标签的分布信息。
在实际中，一个视频可以包括一个或多个视频标签(Tag及hashtag等)，可以获取视频序列中每个视频标识对应的视频标签，并将每个视频的视频标签组成视频标签序列，视频标签序列可以表示为T＝{{t 11,…,t 1n},{t 21,…,t 2n},…{t k1,…,t kn}}，其中，{t 11,…,t 1n},{t 21,…,t 2n},…{t k1,…,t kn}分别表示不同的视频对应的视频标签集合。
获得目标账户历史操作的视频标签序列以后，可以结合每个视频的操作信息，确定每个视频标签的分布信息。
在一种实施方式中，步骤220中根据操作信息确定每个视频标签的分布信息的步骤，可以包括如下步骤：
针对每个视频标签，统计该视频标签出现正向操作信息的第一次数以及出现负向操作信息的第二次数，作为所述分布信息。
在实际中，一个视频标签可能同时具有正向操作信息和负向操作信息，例如，用户点赞了视频v 1，踩了视频v 2，而这两个视频都有标签t 1。得到视频标签序列以后，可以以视频标签为单位，统计每个视频标签出现正向操作信息的第一次数以及出现负向操作信息的第二次数，例如，用户点赞了视频v 1和v 2，踩了视频v 3，视频v 1包括标签t 1、t 2和t 3，视频v 2包括t 2和t 4，视频v 3包括t 2和t 3，则对于标签t 1而言，出现正向操作信息的第一次数为1，出现负向操作信息的第二次数为0；对于标签t 2而言，出现正向操作信息的第一次数为2，出现负向操作信息的第二次数为1；对于标签t 3而言，出现正向操作信息的第一次数为1，出现负向操作信息的第二次数为1；对于标签t 4而言，出现正向操作信息的第一次数为1，出现负向操作信息的第二次数为0。
视频标签出现正向操作信息的第一次数以及出现负向操作信息的第二次数可以作为该视频标签的分布信息。
步骤230，根据所述分布信息，确定对应视频标签的标签权重。
得到每个视频标签的分布信息以后，可以根据该视频标签的正向操作信息和负向操作信息的分布，确定该视频标签的标签权重W′＝{w 1,w 2,…,w m}。
在一种实施方式中，步骤230可以包括如下步骤：
步骤230-1，根据所述第一次数以及所述第二次数计算对应视频标签的权重参考值。
在一种实施方式中，根据第一次数以及第二次数，可以采用如下方式计算对应视频标签的权重参考值：
权重参考值＝(k1+1)/(k1+k2+1)；其中，k1为第一次数，k2为第二次数。
步骤230-2，若所述权重参考值大于第一预设值，则将该视频标签确定为正向视频标签，并将所述视频标签的标签权重确定为所述权重参考值。
步骤230-3，若所述权重参考值小于第二预设值，则将该视频标签确定为负向视频标签，并将所述视频标签的标签权重确定为所述权重参考值的负数，其中，所述第二预设值小于所述第一预设值。
在一种实施例中，在将该视频标签确定为正向视频标签之后，本实施例还可以包括：将所述正向视频标签添加到正向视频标签集合中。
在另一种实施例中，在将该视频标签确定为负向视频标签之后，本实施例还可以包括：将所述负向视频标签添加到负向视频标签集合中。
得到视频标签的权重参考值以后，如果该权重参考值大于第一预设值，则可以将该视频标签确定为正向视频标签，并将该视频标签加入正向视频标签集合中，同时，将该视频标签的标签权重确定为该权重参考值。如果该权重参考值小于第二预设值，则可以将该视频标签确定为负向视频标签，并将该视频标签加入负向视频标签集合中，同时，将该视频标签的标签权重确定为上述权重参考值的负数。其中，第二预设值小于第一预设值。
本实施例将视频标签划分为正向视频标签集合以及负向视频标签集合，正向视频标签集合中的视频标签偏向于正向行为，例如，点赞、完播、分享等；负向视频标签集合中的视频标签偏向于负向行为，例如，踩、倒拇指等，此种划分方式可以便于后续确定视频标签的提权得分为正还是负，从而在视频推荐过程中决定采用提权或者降权处理策略。
步骤240，针对每个视频标签，将所述视频标签的标签权重乘以预设的提权系数得到该视频标签的提权得分。
在该实施例中，可以根据经验设定一个提权系数，例如，将提权系数设定为1-5之间的常数，将每个视频标签的标签权重乘以预设提权系数，可以得到该视频标签的提权得分。其中，对于正向视频标签集合中的视频标签，由于其标签权重为正数，则对应的提权得分也是正数；对于负向视频标签集合中的视频标签，由于其标签权重为负数，则对应的提权得分也是负数
步骤250，对于候选集中每个候选视频，计算所述候选视频包含的视频标签的提权得分之和，得到该候选视频的视频权重。
在该步骤中，候选集为推荐装置按照通用的方法为目标账户初步筛选的多个候选视频组成的集合。该候选集的内容往往较宽泛，即包含较多用户感兴趣的候选视频，但由于坑位资源有限，需要从中筛选出实时性最优的内容。
对于候选集中每个候选视频，可以获取该候选视频携带的视频标签，然后遍历该候选视频的每个视频标签，判断每个视频标签是否具有对应的提权得分，如果视频标签有提权得分，则将获取该提权得分；如果一个视频标签没有提权得分，则可以将该视频标签的提权得分设置为数值0。然后计算该候选视频的所有视频标签的提权得分之和，得到该候选视频的视频权重。
例如，候选视频1携带标签t 1和t 2，其中，标签t 1的提权得分为0.5，标签t 2的提权得分为0.7，则该候选视频的视频权重为0.5+0.7＝1.2；又如，候选视频2携带标签t 1和t 5，其中，标签t 1的提权得分为0.5，标签t 5的提权得分为-2，则该候选视频的视频权重为0.5+(-2)＝-1.5，视频权重为负数时，可以确保用户不喜欢的视频能够快速退出下发，避免负向兴趣点视频的持续下发。
步骤260，根据每个候选视频的视频权重，从所述候选集中选取目标视频进行推荐。
在一种实现中，当确定候选集中每个候选视频的视频权重以后，则可以将候选集中多个候选视频按照视频权重进行排序，然后选取视频权重最大的一个或多个候选视频作为目标视频，并将目标视频推荐下发至客户端。
步骤270，检测已下发的视频中满足设定条件的视频标签，按照该满足设定条件的视频标签的标签权重，对所述候选集中包含该满足设定条件的视频标签的、未下发的候选视频进行降权处理。
在一种实施方式中，可以统计每个视频标签对应的已下发视频的数量，当一视频标签对应的已下发视频的数量达到预设数量阈值时，则可以判定该视频标签满足设定条件，则可以对候选集中包含该视频标签的、未下发的视频进行降权处理，以限制后续对该视频标签的视频下发量，避免满屏全是单一兴趣点视频的情况。
在一种实施方式中，步骤270可以包括如下步骤：
将所述候选集中包含所述满足设定条件的视频标签的、未下发的视频的视频权重减去所述满足设定条件的视频标签的提权得分，得到所述候选视频的新的视频权重。
在该实施例中，降权的实现过程可以为，对于候选集中未下发的候选视频，可以判断上述未下发的候选视频是否存在满足设定条件的视频标签，如果上述未下发的候选视频存在满足设定条件的视频标签，则对存在满足设定条件的视 频标签的候选视频进行降权处理，即将该候选视频的视频权重减去满足设定条件的视频标签的提权得分，得到该候选视频的新的视频权重，从而实现对候选视频的视频权重的动态调整。
例如，假设目标账户历史操作的视频的视频标签有三个，分别是A、B、C，对应的提权得分依次为1分、0.5分和0.1分。那么，可以对候选集中所有存在视频标签A的视频(此处称为A类视频)提权1分，对候选集中所有存在视频标签B的视频(此处称为B类视频)提权0.5分，对候选集中所有存在视频标签C的视频(此处称为C类视频)提权0.1分。随着坑位不断填补，当A类视频下发满足需求后，则取消对候选集中剩余A类视频的提权，即对剩余A类视频降权1分；当B类视频下发满足需求后，则取消对候选集中剩余B类视频的提权，即对剩余B类视频降权0.5分；当C类视频下发满足需求后，则取消对候选集中剩余C类视频的提权，即对剩余C类视频降权0.1分。经过以上动态提权过程，能大大提高A、B、C三个兴趣点视频的下发概率。通过对超出一定数量的兴趣点视频进行降权操作，避免满屏全是单一兴趣点视频的情况，从而满足用户的兴趣需求。
为了实时捕捉用户不断变化的兴趣偏好，本实施例根据目标账户短期的历史行为数据中的视频标识对应的视频所携带的视频标签，确定每个视频标签的标签权重，然后根据标签权重来确定视频标签实时的提权得分，基于每个视频标签的提权得分可以计算候选集中每个候选视频的视频权重，确保用户实时兴趣点视频能够及时下发，提高视频下发的概率。然后在检测到已下发的视频中满足设定条件的视频标签时，则可以按照该视频标签的标签权重，对候选集中包含该视频标签的、未下发的视频进行降权处理，以此动态调整候选集中各候选视频的视频权重。
实施例三
图3是本申请实施例三提供的一种视频推荐装置实施例的结构框图，该视频推荐装置可以位于服务器中，可以包括如下模块：
标签权重确定模块310，设置为获取目标账户历史操作的视频标识，并确定所述视频标识对应的视频中包含的多个视频标签中各每个视频标签的标签权重；视频权重确定模块320，设置为基于每个视频标签的标签权重，确定预设的候选集中多个候选视频中每个候选视频的视频权重；视频推荐模块330，设置为根据每个候选视频的视频权重，从所述候选集中选取目标视频进行推荐；降权处理模块340，设置为检测已下发的视频中满足设定条件的视频标签，按照所述满足设定条件的视频标签的标签权重，对所述候选集中包含所述满足设定条件的视 频标签的、未下发的候选视频进行降权处理。
本申请实施例所提供的上述视频推荐装置可执行本申请任意实施例所提供的视频推荐方法，具备执行方法相应的功能模块和效果。
实施例四
图4是本申请实施例四提供的一种电子设备的结构示意图，如图4所示，该电子设备包括处理器410、存储器420、输入装置430和输出装置440；电子设备中处理器410的数量可以是一个或多个，图4中以一个处理器410为例；电子设备中的处理器410、存储器420、输入装置430和输出装置440可以通过总线或其他方式连接，图4中以通过总线连接为例。
存储器420作为一种计算机可读存储介质，可设置为存储软件程序、计算机可执行程序以及模块，如本申请实施例中的视频推荐方法对应的程序指令/模块。处理器410通过运行存储在存储器420中的软件程序、指令以及模块，从而执行电子设备的多种功能应用以及数据处理，即实现上述的视频推荐方法。
实施例五
本申请实施例五还提供一种包含计算机可执行指令的存储介质，所述计算机可执行指令在由服务器的处理器执行时用于执行实施例一至实施例二中任一实施例中的视频推荐方法。
对于装置、电子设备、存储介质实施例而言，由于其与方法实施例基本相似，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。

Claims (15)

  1. 一种视频推荐方法，包括：
    获取目标账户历史操作的视频标识，并确定所述视频标识对应的视频中包含的多个视频标签中每个视频标签的标签权重；
    基于每个视频标签的标签权重，确定预设的候选集中多个候选视频中每个候选视频的视频权重；
    根据每个候选视频的视频权重，从所述候选集中选取目标视频进行推荐；
    检测已下发的视频中满足设定条件的视频标签，按照所述满足设定条件的视频标签的标签权重，对所述候选集中包含所述满足设定条件的视频标签且未下发的候选视频进行降权处理。
  2. 根据权利要求1所述的方法，其中，所述获取目标账户历史操作的视频标识，并确定所述视频标识对应的视频中包含的多个视频标签中每个视频标签的标签权重，包括：
    获取所述目标账户的历史行为数据，其中，所述历史行为数据包括所述目标账户操作的多个视频标识以及操作信息；
    获取所述多个视频标识对应的视频所包含的所述多个视频标签，并根据所述操作信息确定每个视频标签的分布信息；
    根据所述分布信息，确定所述每个视频标签的标签权重。
  3. 根据权利要求2所述的方法，其中，所述操作信息包括正向操作信息以及负向操作信息；
    所述根据所述操作信息确定每个视频标签的分布信息，包括：
    针对每个视频标签，统计所述每个视频标签出现正向操作信息的第一次数以及出现负向操作信息的第二次数，作为所述分布信息。
  4. 根据权利要求3所述的方法，其中，所述根据所述分布信息，确定所述每个视频标签的标签权重，包括：
    根据所述第一次数以及所述第二次数计算所述每个视频标签的权重参考值；
    在所述权重参考值大于第一预设值的情况下，将所述每个视频标签确定为正向视频标签，并将所述每个视频标签的标签权重确定为所述权重参考值；
    在所述权重参考值小于第二预设值的情况下，将所述每个视频标签确定为负向视频标签，并将所述每个视频标签的标签权重确定为所述权重参考值的负数；
    其中，所述第二预设值小于所述第一预设值。
  5. 根据权利要求4所述的方法，在所述将所述每个视频标签确定为正向视频标签之后，还包括：
    将所述正向视频标签添加到正向视频标签集合中；
    在所述将所述每个视频标签确定为负向视频标签之后，还包括：
    将所述负向视频标签添加到负向视频标签集合中。
  6. 根据权利要求4所述的方法，其中，所述根据所述第一次数以及所述第二次数计算所述每个视频标签的权重参考值，包括：
    所述权重参考值＝(所述第一次数+1)/(所述第一次数+所述第二次数+1)。
  7. 根据权利要求2-6中任一项所述的方法，其中，所述历史行为数据为所述目标账户最近一次会话的行为数据。
  8. 根据权利要求1-6中任一项所述的方法，其中，所述基于每个视频标签的标签权重，确定预设的候选集中多个候选视频中每个候选视频的视频权重，包括：
    针对每个视频标签，将所述每个视频标签的标签权重乘以预设的提权系数得到所述每个视频标签的提权得分；
    对于候选集中每个候选视频，计算所述候选视频包含的视频标签的提权得分之和，得到所述每个候选视频的视频权重。
  9. 根据权利要求8所述的方法，其中，所述按照所述满足设定条件的视频标签的标签权重，对所述候选集中包含所述满足设定条件的视频标签且未下发的候选视频进行降权处理，包括：
    将所述候选集中包含所述满足设定条件的视频标签且未下发的候选视频的视频权重减去所述满足设定条件的视频标签的提权得分，得到所述候选视频的新的视频权重。
  10. 根据权利要求1-6中任一项所述的方法，其中，所述检测已下发的视频中满足设定条件的视频标签，包括：
    统计每个视频标签对应的已下发视频的数量；
    在一视频标签的已下发视频的数量达到预设数量阈值的情况下，判定所述一视频标签满足所述设定条件。
  11. 根据权利要求1所述的方法，其中，所述根据每个候选视频的视频权重，从所述候选集中选取目标视频进行推荐，包括：
    将所述候选集中所述多个候选视频按照所述视频权重进行排序；
    选取视频权重最大的前至少一个候选视频作为所述目标视频。
  12. 根据权利要求1所述的方法，在所述获取目标账户历史操作的视频标识之前，还包括：
    接收客户端发送的视频拉取请求，其中，所述视频拉取请求包括所述目标账户。
  13. 一种视频推荐装置，包括：
    标签权重确定模块，设置为获取目标账户历史操作的视频标识，并确定所述视频标识对应的视频中包含的多个视频标签中每个视频标签的标签权重；
    视频权重确定模块，设置为基于每个视频标签的标签权重，确定预设的候选集中多个候选视频中每个候选视频的视频权重；
    视频推荐模块，设置为根据每个候选视频视频权重，从所述候选集中选取目标视频进行推荐；
    降权处理模块，设置为检测已下发的视频中满足设定条件的视频标签，按照所述满足设定条件的视频标签的标签权重，对所述候选集中包含所述满足设定条件的视频标签且未下发的候选视频进行降权处理。
  14. 一种电子设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，其中，所述处理器执行所述程序时实现如权利要求1-12中任一项所述的视频推荐方法。
  15. 一种计算机可读存储介质，存储有计算机程序，其中，所述程序被处理器执行时实现如权利要求1-12中任一项所述的视频推荐方法。
PCT/CN2021/115027 2020-09-29 2021-08-27 视频推荐方法及装置 WO2022068492A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP21874153.6A EP4224874A4 (en) 2020-09-29 2021-08-27 METHOD AND APPARATUS FOR VIDEO RECOMMENDATION
US18/246,004 US20230362423A1 (en) 2020-09-29 2021-08-27 Method and electronic device for recommending videos, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011052701.1 2020-09-29
CN202011052701.1A CN112188295B (zh) 2020-09-29 2020-09-29 一种视频推荐方法及装置

Publications (1)

Publication Number Publication Date
WO2022068492A1 true WO2022068492A1 (zh) 2022-04-07

Family

ID=73946605

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/115027 WO2022068492A1 (zh) 2020-09-29 2021-08-27 视频推荐方法及装置

Country Status (4)

Country Link
US (1) US20230362423A1 (zh)
EP (1) EP4224874A4 (zh)
CN (1) CN112188295B (zh)
WO (1) WO2022068492A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116233554A (zh) * 2023-01-17 2023-06-06 北京达佳互联信息技术有限公司 视频播放方法、装置、电子设备和存储介质

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112188295B (zh) * 2020-09-29 2022-07-05 有半岛(北京)信息科技有限公司 一种视频推荐方法及装置
CN112784111B (zh) * 2021-03-12 2024-07-02 有半岛(北京)信息科技有限公司 视频分类方法、装置、设备及介质
US20240163515A1 (en) * 2021-03-30 2024-05-16 Boe Technology Group Co., Ltd. Method and device for recommending real-time audios and/or videos, and computer storage medium
CN114419501A (zh) * 2022-01-11 2022-04-29 平安普惠企业管理有限公司 视频推荐方法、装置、计算机设备及存储介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103440335A (zh) * 2013-09-06 2013-12-11 北京奇虎科技有限公司 视频推荐方法及装置
US20170060870A1 (en) * 2015-08-24 2017-03-02 Google Inc. Video recommendation based on video titles
CN106791963A (zh) * 2016-12-08 2017-05-31 Tcl集团股份有限公司 一种电视节目推荐方法及系统
US20170171334A1 (en) * 2015-12-14 2017-06-15 Le Holdings (Beijing) Co., Ltd. Single-account multiple-preference recommendation method for video website and electronic device
CN109657138A (zh) * 2018-12-10 2019-04-19 深圳墨世科技有限公司 一种视频推荐方法、装置、电子设备及存储介质
CN110059221A (zh) * 2019-03-11 2019-07-26 咪咕视讯科技有限公司 视频推荐方法、电子设备及计算机可读存储介质
CN112188295A (zh) * 2020-09-29 2021-01-05 有半岛(北京)信息科技有限公司 一种视频推荐方法及装置

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8489515B2 (en) * 2009-05-08 2013-07-16 Comcast Interactive Media, LLC. Social network based recommendation method and system
CN102957950A (zh) * 2012-07-23 2013-03-06 华东师范大学 一种用于视频推荐的用户隐式评分方法
CN105512331B (zh) * 2015-12-28 2019-03-26 海信集团有限公司 一种视频推荐方法及装置
CN106294830A (zh) * 2016-08-17 2017-01-04 合智能科技（深圳）有限公司 多媒体资源的推荐方法及装置
KR20190007816A (ko) * 2017-07-13 2019-01-23 삼성전자주식회사 동영상 분류를 위한 전자 장치 및 그의 동작 방법
CN111279709B (zh) * 2018-08-10 2022-11-08 微软技术许可有限责任公司 提供视频推荐
CN109729395B (zh) * 2018-12-14 2022-02-08 广州市百果园信息技术有限公司 视频质量评估方法、装置、存储介质和计算机设备
CN110287372B (zh) * 2019-06-26 2021-06-01 广州市百果园信息技术有限公司 用于负反馈的标签确定方法、视频推荐方法及其装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103440335A (zh) * 2013-09-06 2013-12-11 北京奇虎科技有限公司 视频推荐方法及装置
US20170060870A1 (en) * 2015-08-24 2017-03-02 Google Inc. Video recommendation based on video titles
US20170171334A1 (en) * 2015-12-14 2017-06-15 Le Holdings (Beijing) Co., Ltd. Single-account multiple-preference recommendation method for video website and electronic device
CN106791963A (zh) * 2016-12-08 2017-05-31 Tcl集团股份有限公司 一种电视节目推荐方法及系统
CN109657138A (zh) * 2018-12-10 2019-04-19 深圳墨世科技有限公司 一种视频推荐方法、装置、电子设备及存储介质
CN110059221A (zh) * 2019-03-11 2019-07-26 咪咕视讯科技有限公司 视频推荐方法、电子设备及计算机可读存储介质
CN112188295A (zh) * 2020-09-29 2021-01-05 有半岛(北京)信息科技有限公司 一种视频推荐方法及装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4224874A4

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116233554A (zh) * 2023-01-17 2023-06-06 北京达佳互联信息技术有限公司 视频播放方法、装置、电子设备和存储介质

Also Published As

Publication number Publication date
US20230362423A1 (en) 2023-11-09
CN112188295A (zh) 2021-01-05
EP4224874A4 (en) 2024-03-06
CN112188295B (zh) 2022-07-05
EP4224874A1 (en) 2023-08-09

Similar Documents

Publication Publication Date Title
WO2022068492A1 (zh) 视频推荐方法及装置
US8843470B2 (en) Meta classifier for query intent classification
CN108304490B (zh) 基于文本的相似度确定方法、装置及计算机设备
CN104679818B (zh) 一种视频关键帧提取方法及系统
CN108762907B (zh) 基于多个客户端的任务处理方法及系统
US20170177584A1 (en) Media names matching and normalization
WO2021000830A1 (zh) 人脸识别的方法、装置、计算机设备和存储介质
CN109120949B (zh) 视频集合的视频消息推送方法、装置、设备及存储介质
CN115658282A (zh) 服务器算力管理分配方法、系统、网络设备和存储介质
CN105893390A (zh) 一种应用程序的处理方法及电子设备
CN109376288A (zh) 一种实现语义搜索的云计算平台及其均衡方法
CN105100277A (zh) 一种软件巡检方法、装置及系统
CN107515876B (zh) 一种特征模型的生成、应用方法及装置
Finlay World Heavyweight Championship boxing: The past 30+ years of the male division
CN111611973A (zh) 目标用户识别的方法、装置及存储介质
WO2021240679A1 (ja) 映像処理装置、映像処理方法、及び、記録媒体
US20130091131A1 (en) Meta-model distributed query classification
CN110139160A (zh) 一种预测系统及方法
CN115412737A (zh) 直播回源中继节点确定方法及装置
CN104618492A (zh) 一种基于多个服务器的数据处理系统
CN113326829B (zh) 视频中手势的识别方法、装置、可读存储介质及电子设备
CN107577808A (zh) 一种多级列表页排序的方法、装置、服务器及介质
CN111143688B (zh) 一种基于移动新闻客户端的评估方法及其系统
RU2809340C1 (ru) Способ, устройство и электронное устройство для рекомендации видео и носитель данных
CN115514951A (zh) 一种用于评估播放质量的方法、装置、介质及计算机设备

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: 21874153

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2023108391

Country of ref document: RU

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2021874153

Country of ref document: EP

Effective date: 20230502