WO2022007626A1 - 视频内容推荐方法、装置及计算机设备 - Google Patents

视频内容推荐方法、装置及计算机设备 Download PDF

Info

Publication number
WO2022007626A1
WO2022007626A1 PCT/CN2021/101626 CN2021101626W WO2022007626A1 WO 2022007626 A1 WO2022007626 A1 WO 2022007626A1 CN 2021101626 W CN2021101626 W CN 2021101626W WO 2022007626 A1 WO2022007626 A1 WO 2022007626A1
Authority
WO
WIPO (PCT)
Prior art keywords
video
target
attribute
recommendation score
sequence
Prior art date
Application number
PCT/CN2021/101626
Other languages
English (en)
French (fr)
Inventor
吴俊豪
谢鹏
Original Assignee
上海哔哩哔哩科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 上海哔哩哔哩科技有限公司 filed Critical 上海哔哩哔哩科技有限公司
Priority to US18/014,339 priority Critical patent/US20230300417A1/en
Publication of WO2022007626A1 publication Critical patent/WO2022007626A1/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]
    • H04N21/40—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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/73—Querying
    • G06F16/732—Query formulation
    • 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]
    • H04N21/80—Generation or processing of content or additional data by content creator independently of the distribution process; Content per se
    • H04N21/83—Generation or processing of protective or descriptive data associated with content; Content structuring
    • H04N21/845—Structuring of content, e.g. decomposing content into time segments
    • H04N21/8456—Structuring of content, e.g. decomposing content into time segments by decomposing the content in the time domain, e.g. in time segments

Definitions

  • the present application relates to the field of Internet technologies, and in particular, to a video content recommendation method, apparatus, and computer equipment.
  • Video platforms generally also provide content recommendation functions to increase the user stickiness of video platforms.
  • the inventor realized that most of the existing video platforms generally have a video recommendation system that retrieves a certain amount of candidate sets from the video library according to the user's personalized interests, then scores and sorts these candidate sets, and then recommends to the corresponding client. The more times a user clicks on a certain type of video or the longer the viewing time, the more recommended videos of this type will be obtained, and the user's click on this type of recommended video will further strengthen the recommendation of this type of video.
  • the present application proposes a video content recommendation method, device and computer equipment, which can solve the problems of lack of diversity and poor recommendation flexibility in the prior art in the video content recommendation process.
  • the present application provides a method for recommending video content, the method comprising:
  • the acquiring the target attribute of the first video includes: when the target attribute of the first video is empty, marking the first video with a preset video attribute, and then converting the video attribute as the target attribute of the first video.
  • modifying the initial recommendation score of the first video according to the n and the sorting distance between each target video and the first video, to obtain a modified recommendation score includes: when When the n is greater than or equal to the preset threshold N, the revised recommendation score of the first video is set to 0.
  • modifying the initial recommendation score of the first video according to the n and the sorting distance between each target video and the first video, to obtain a modified recommendation score includes: when When the n is less than the preset threshold N, obtain the position number i of the target video in the video sequence section and the position number k of the first video in the video sequence section; according to the i, the k and a preset correction formula correct the initial recommendation score.
  • the target attribute includes a tag attribute and an author attribute
  • the correction formula includes:
  • scorek is the initial recommendation score of the first video
  • score'k is the revised recommendation score of the first video
  • count(tagsk) is the number of tag attributes of the first video
  • count(tagsi) is The number of tag attributes of the target video
  • demo(distance(i, k), tag) is the first correction function
  • demo(distance(i, k), up) is the second correction function
  • tag is the tag attribute
  • up is the author attribute.
  • the first correction function and the second correction function are decay functions whose value range is [0, 1], including a linear function or a quadratic function.
  • the correction formula when the first correction function and the second correction function are both half-life functions, the correction formula includes:
  • Ttag is the decay constant preset according to the tag attribute
  • Tup is the decay constant preset according to the author attribute.
  • modifying the initial recommendation score of the first video according to the n and the sorting distance between each target video and the first video, to obtain a modified recommendation score includes: The user terminal to be recommended obtains the corresponding target user attribute; according to the n and the sorting distance between each target video and the first video, in combination with the target user attribute and/or the video type of the first video , revising the initial recommendation score of the first video to obtain a revised recommendation score corresponding to the target user.
  • the present application also provides a video content recommendation device, the device comprising:
  • the pre-adding module is used to sequentially select the video to be selected from the video set to be selected as the first video to be pre-added to the last sequence of the recommended video sequence;
  • the interception module is used to intercept the video sequence that contains the first video in the recommended video sequence.
  • the video sequence section of the preset number of videos; the acquisition module is used to obtain the target attribute and the initial recommendation score of the first video; the search module is used to find out and remove the first video in the video sequence section The target video with the target attribute other than the video, and the number n of the target video is counted; the correction module is used for pairing according to the n and the sorting distance between each target video and the first video
  • the initial recommendation score of the first video is revised to obtain a revised recommendation score;
  • a selection module is configured to put the first video back into the candidate video set; and select a revision from the candidate video set
  • the video with the highest recommendation score is added to the last sequence of the video sequence to be recommended.
  • the present application also proposes a computer device, the computer device includes a memory and a processor, the memory stores computer-readable instructions that can be executed on the processor, and the computer-readable instructions are executed by the processor. The following steps are implemented when the processor described above is executed:
  • the present application also provides a computer-readable storage medium, where the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to make The at least one processor performs the following steps:
  • the video content recommendation method, device, computer equipment, and computer-readable storage medium proposed in the present application can pre-add the candidate video of the candidate video set as the first video to the recommended video sequence, and then intercept the video containing the first video. the video sequence section, and then according to the position sequence number of the first video in the video sequence section and the position sequence number of other videos with the target attribute of the first video in the video sequence section
  • the recommendation score is revised to obtain a revised recommendation score, and finally a final candidate video is selected and added to the recommended video sequence according to the revised recommendation score.
  • the negative influence of all similar videos is accumulated according to the similar videos to achieve diversity control. Therefore, the diversity of the recommended content is improved, and the flexibility of the recommendation method is also improved.
  • FIG. 1 is a schematic diagram of an application environment of an embodiment of the present application.
  • FIG. 2 is a schematic flowchart of a specific embodiment of a video content recommendation method of the present application
  • FIG. 3 is a schematic flowchart of a specific embodiment of step S208 in FIG. 2;
  • FIG. 4 is an effect diagram of a video content recommendation method according to a specific embodiment of the present application.
  • FIG. 5 is a schematic diagram of a program module of an embodiment of an apparatus for recommending video content of the present application
  • FIG. 6 is a schematic diagram of an optional hardware architecture of the computer device of the present application.
  • the video recommendation method in the prior art easily leads to the continuous or high-frequency appearance of videos of the same type, thereby reducing the user's willingness to watch and affecting the user's use experience. In other words, it is difficult for existing video recommendation systems to ensure the diversity of recommended videos, and the recommendation flexibility is lacking.
  • FIG. 1 is a schematic diagram of an application environment of an embodiment of the present application.
  • the computer device 1 is connected to a user terminal.
  • the computer device 1 can obtain user information and user behavior records on the user terminal, and the user information includes information such as user gender, occupation, age, and region; the user behavior records include user viewing records, favorite records, click records, click records, etc. Like records and other interactive records.
  • the computer device 1 analyzes the points of interest of the user corresponding to the user terminal according to the user information and user behavior records of the user terminal. Or the number of likes is more, so as to analyze the user's preference for videos of this type of video.
  • the computer device 1 selects a preset number of videos from a video resource server or database (not shown in FIG. 1 ) according to the user's point of interest as a candidate video set.
  • the computer device 1 will select other popular videos in addition to adding videos that match the user's point of interest to the candidate video set, for example, according to the like rate/follow rate/interaction rate/click of the video Popular videos are filtered out and added to the candidate video set for subsequent recommendation to the user terminal corresponding to the user.
  • the user terminal can be used as a mobile phone, a tablet, a portable device, a PC, or other electronic devices with a display function;
  • the computer device 1 can be used as a PC, a data server, and the like.
  • FIG. 2 is a schematic flowchart of an embodiment of a video content recommendation method of the present application. It can be understood that the flowchart in this embodiment of the method is not used to limit the sequence of executing steps. The following description will be exemplified by taking the computer device 1 as the execution subject.
  • the video content recommendation method may include steps S200-S206.
  • Step S200 sequentially selecting the videos to be selected from the video set to be selected as the first video to be pre-added to the last sequence of the recommended video sequence.
  • Step S202 cutting out a video sequence section including a preset number of videos of the first video from the recommended video sequence.
  • Step S204 acquiring the target attribute and initial recommendation score of the first video.
  • the computer device 1 filters out a preset number of videos according to the user's point of interest, and at the same time selects a certain number of popular videos to be added to the video set to be selected together.
  • the computer device 1 sequentially selects the candidate videos from the candidate video set as the first video and pre-adds them to the last sequence of the recommended video sequence.
  • pre-adding is only temporary, and the computer device 1 sequentially pre-adds the videos to be selected in the video set to be selected to the recommended video sequence, so that the pre-added videos to the recommended video sequence can be sequentially added to the recommended
  • the candidate videos in the video sequence undergo subsequent correction of recommendation scores.
  • the computer device 1 pre-adds the candidate video in the candidate video set as the first video to the recommended video sequence, then further from the first video, to the sequence of the recommended video sequence Cuts a video sequence segment of a preset number of videos in the reverse direction.
  • the computer device 1 acquires the target attribute and the initial recommendation score of the first video.
  • the target attribute includes a tag attribute and an author attribute.
  • the video will generally be tagged with a tag attribute, such as an animation tag and a funny video tag. , entertainment variety show tags, etc.; then the video uploaded to the video data server will also automatically carry the information of the user who uploaded the video, that is, the author attribute.
  • the videos in the candidate video set also include an initial recommendation score
  • the initial recommendation score can be understood as when the computer device 1 adds the video to the candidate video set, the video is evaluated recommended weight value.
  • the recommendation weight value may be calculated according to the like rate/attention rate/interaction rate/click rate of the video, or given by the evaluation of viewing users, or by manually playing the video Comprehensive evaluation of the situation and reviews.
  • the computer device 1 Since the computer device 1 selects a certain number of videos from the video data server as the video set to be selected, the computer device 1 can obtain the target attribute of the first video; Initial recommendation score.
  • the target attribute includes any attribute extracted according to relevant information of the video, the relevant information including but not limited to the classification, introduction, author information, uploader information, comment information, Barrage information, etc.
  • the target attribute may be determined by analyzing the above-mentioned related information and extracting keywords in the above-mentioned information.
  • the target attribute includes, but is not limited to, tag data and/or author attribute of the target video.
  • the computer device 1 may determine the tag attribute of the first video according to the video section, title, introduction, author information, etc. of the first video; the computer device 1 may determine the tag attribute of the first video according to the Author information, such as the division of the author, the number of fans, the classification of the historically uploaded videos, the keywords of the historically uploaded videos, gender, age, hobbies, etc., determine the author attribute.
  • the computer device 1 when the computer device acquires the target attribute of the first video, the target attribute of the first video is empty, then the computer device 1 will mark the first video
  • the preset video attributes are then used as the target attributes of the first video. For example, when the first video is not tagged when the user uploads the video, the computer device 1 cannot acquire the tag attribute of the first video, or the acquired tag attribute is an invalid tag, such as , does not belong to the default canonical label, or is displayed in garbled characters. Then, the computer device 1 will add a preset tag attribute to the first video.
  • Step S206 Find out the target videos with the target attribute except the first video in the video sequence section, and count the number n of the target videos.
  • Step S208 correcting the initial recommendation score of the first video according to the n and the sorting distance between each target video and the first video, to obtain a revised recommendation score.
  • the computer device 1 after obtaining the target attribute of the first video, the computer device 1 further obtains attribute values of other videos in the video sequence section, and then finds a video with the target attribute as the target video , and count the number n of all target videos.
  • the computer device 1 presets rules for video recommendation. For example, in order to relieve the fatigue of users watching videos, in a preset section of the recommended video sequence, no more than a certain number of identical target property. Therefore, after counting the number n of the target videos in the video sequence section, the computer device 1 further determines whether the n is greater than or equal to a preset threshold N.
  • the computer device 1 sets the revised recommendation score of the first video to 0. For example, in the video sequence section, the videos with the same tag attribute cannot exceed half, if more than half, for example, 10 or more than 5 videos are tagged with animation, then the computer device 1 will pre- The recommendation score of the added first video with an anime tag is set as 0 points. In this way, videos with the same target attribute appearing in succession can be restricted, thereby reducing the continuous or high-frequency appearance of videos of the same type.
  • the initial order of the first video is determined according to the n and the sorting distance between each target video and the first video.
  • the recommendation score is revised, and the revised recommendation score is obtained, including:
  • Step S300 Obtain the position sequence number i of the target video in the video sequence section and the position sequence number k of the first video in the video sequence section.
  • Step S302 correcting the initial recommendation score according to the i, the k and a preset correction formula.
  • the target attribute includes a tag attribute and an author attribute
  • the correction formula includes:
  • scorek is the initial recommendation score of the first video
  • score'k is the revised recommendation score of the first video
  • count(tagsk) is the number of tag attributes of the first video
  • count(tagsi) is The number of tag attributes of the target video
  • demo(distance(i, k), tag) is the first correction function
  • demo(distance(i, k), up) is the second correction function
  • tag is the tag attribute
  • up is the author attribute.
  • the first correction function and the second correction function are decay functions whose value range is [0, 1], including a linear function or a quadratic function.
  • the first modification function is used to modify the initial recommendation score of the first video according to the difference between the target video of the same tag attribute and the position number of the first video;
  • the second modification The function is used to revise the initial recommendation score of the first video according to the difference between the position serial numbers of the target video with the same author attribute and the first video.
  • the correction formula when the first correction function and the second correction function are both half-life functions, the correction formula includes:
  • Ttag is the decay constant preset according to the tag attribute
  • Tup is the decay constant preset according to the author attribute.
  • the Ttag and the Tup are an attenuation parameter obtained through offline evaluation or online experiments, specifically: according to the initial recommendation score of the video in the video set to be selected, user information, user behavior records And the video information in the said candidate video set, such as the distribution of various label attributes included in all videos, the probability of consecutively appearing videos of the same label/video author, and the probability that the interval between similar videos is 1/2/3/x, The probability of similar videos appearing in 5/6/x consecutive videos, the change in the proportion of video categories watched by each user, and the historical interaction rate indicators of the selected videos and other information to set the Ttag and the Tup offline; or Create a layer of diversity-controlled experiments through the preset online recommendation engine, test the pros and cons of the online recommendation indicators and the retained data to evaluate each group of weight reduction parameters according to the experimental access traffic, so as to select the optimal Ttag. and the Tup.
  • the computer device 1 may also obtain the corresponding target user attribute from the user terminal to be recommended, and then according to the n and the sorting distance between each target video and the first video, Combining the attributes of the target user and/or the video type of the first video, the initial recommendation score of the first video is revised to obtain a revised recommendation score corresponding to the target user.
  • the target user attribute and the video type of the first video may correspond to different correction parameters respectively, so as to correct the initial recommendation score, for example, for different target user attributes And different video types, the initial recommendation score of the video can be directly weighted or down-weighted; or, in another embodiment, since each user has different acceptance levels for continuously watching the same type of video content, therefore,
  • the target user attribute of the target user to be recommended can be obtained first, and the target user attribute includes the user's attenuation constants Ttag and Tup for different types of video content.
  • the computer device 1 can, according to the video type of the first video, Select the corresponding Ttag and Tup from the acquired attributes of the target user, and then combine them into the above-described calculation method of the revised recommendation score, thereby calculating the revised recommendation score corresponding to the target user.
  • Step S210 returning the first video to the candidate video set.
  • Step S212 Select the video with the highest revised recommendation score from the video set to be selected and add it to the last sequence of the video sequence to be recommended.
  • the computer device 1 After obtaining the revised recommendation score of the first video, the computer device 1 puts the first video back into the candidate video set, and when the initial recommendation scores of all videos in the candidate video set are all After the correction is performed in the above manner, the computer device 1 selects the video with the highest correction recommendation score from the candidate video set and adds it to the last sequence of the to-be-recommended video sequence.
  • the computer device 1 pre-adds the videos in the video set to be selected to the recommended video sequence, and then establishes an evaluation effective range/sliding window (that is, the video sequence section) in the recommended video sequence, and then according to the evaluation effective range/ Other videos in the sliding window with the same tag attribute or author attribute correct the initial recommendation score of the pre-added video.
  • the effective range/sliding window can be introduced. For example, similar videos that are far away from the position to be selected can be ignored directly, and at the same time, the time-consuming of online calculation can be saved.
  • FIG. 4 is an effect diagram of a video content recommendation method according to an exemplary embodiment of the present application.
  • the computer device 1 selects the first video from the video set to be selected as the last sequence number in the first video pre-added recommended video sequence, that is, the position of sequence number 4;
  • the video is intercepted as a video sequence section; the target attribute of the first video is obtained, namely tag: 0 and the initial recommendation score score, and then the video with tag: 0 in the video sequence section is found, that is, the video sequence area.
  • the revised recommendation score score' of the first video is calculated, and then the first video is put back into the video set to be selected. Finally, the computer device 1 sequentially corrects the initial recommendation scores of the videos in the video set to be selected according to this method, and then selects the video with the highest revised recommendation score and adds it to the position of serial number 4 of the recommended video sequence.
  • the computer device 1 described in this application can pre-add the candidate video of the candidate video set as the first video to the recommended video sequence, and then intercept the video sequence section including the first video, and then select the video sequence according to the first video.
  • the position sequence number in the video sequence section and the position sequence number of other videos with the target attribute of the first video in the video sequence section are revised to correct the initial recommendation score of the first video, and the revised recommendation score is obtained.
  • the recommendation score selects the final candidate video to add to the recommended video sequence. In this way, the negative influence of all similar videos is accumulated according to the similar videos to achieve diversity control. Therefore, the diversity of the recommended content is improved, and the flexibility of the recommendation method is also improved.
  • this rule is to limit the recommended quantity of video content of the same type. For example, in the process of video content recommendation, among the five videos recommended to users , no more than 2 videos of the same type. Therefore, it is difficult to apply this rule to all users, and it is difficult to balance satisfying the restriction rule and the recommendation index; and the different processing of different types of videos in this rule is based on human experience judgment, and cannot be further improved and extended.
  • video content recommendation is performed according to the MMR (Maximal Marginal Relevance, maximum boundary correlation method) method, in which MMR only selects the video with the greatest similarity to reduce the weight, and the returned result is the weight reduction of similar videos, which is not considered in this method.
  • MMR Maximum Marginal Relevance, maximum boundary correlation method
  • the experience of continuous similar videos is worse than that of spaced similar videos; when waiting for selection, only the selected similar videos are selected with the largest one.
  • Relevant videos have negative influence, that is, weight reduction, without the superposition of multiple similar videos.
  • this application can support more personalized recommendations, has better control over the proportion of recommended video categories, and can obtain better user satisfaction from the perspective of user consumption; and MMR Compared with the method, this application can support the accumulation of negative influence of multiple selected videos of the same type on the candidates, so if some reasons have caused multiple videos of the same type to appear at a higher frequency, the possibility of subsequent videos of the same type will appear faster. It supports more personalized recommendation capabilities, and adopts different processing for different users and different types of videos. To sum up, it can be seen that this application can achieve:
  • this application can comprehensively consider the entire recommended video sequence, the similarity between the video to be selected and all similar videos, and the position distance in the sequence, so as to ensure the index effect (like rate/attention rate/overall interaction rate) of the recommendation system. etc.), and the recommendation online system can complete the entire recommendation calculation with less resource consumption and response time.
  • FIG. 5 schematically shows a block diagram of an apparatus for recommending video content according to Embodiment 2 of the present application.
  • the apparatus for recommending video content may be divided into one or more program modules, and one or more program modules are stored in a storage medium, and executed by one or more processors to complete the embodiments of the present application.
  • the program modules referred to in the embodiments of the present application refer to a series of computer-readable instruction segments capable of performing specific functions. The following description will specifically introduce the functions of each program module in this embodiment.
  • the video content recommendation apparatus 400 may include a pre-adding module 410, a clipping module 420, an obtaining module 430, a searching module 440, a correction module 450 and a selection module 460, wherein:
  • the pre-adding module 410 is used to sequentially select the video to be selected from the video set to be selected as the first video to be pre-added to the last sequence of the recommended video sequence.
  • a clipping module 420 configured to clip a video sequence segment including a preset number of videos of the first video from the recommended video sequence.
  • the obtaining module 430 is configured to obtain the target attribute and the initial recommendation score of the first video.
  • the search module 440 is configured to search out the target video with the target attribute except the first video in the video sequence section, and count the number n of the target video.
  • the modification module 450 is configured to modify the initial recommendation score of the first video according to the n and the sorting distance between each target video and the first video to obtain a modified recommendation score.
  • the selection module 460 is configured to put the first video back into the candidate video set; and select the video with the highest revised recommendation score from the candidate video set and add it to the last sequence of the video sequence to be recommended.
  • the modification module 450 is further configured to: when the n is greater than or equal to a preset threshold N, set the modification recommendation score of the first video to 0. When the n is less than the preset threshold N, obtain the position number i of the target video in the video sequence section and the position number k of the first video in the video sequence section; according to the The initial recommendation score is corrected by the i, the k and the preset correction formula.
  • the target attribute includes tag attribute and author attribute
  • the correction formula includes:
  • scorek is the initial recommendation score of the first video
  • score'k is the revised recommendation score of the first video
  • count(tagsk) is the number of tag attributes of the first video
  • count(tagsi) is The number of tag attributes of the target video
  • demo(distance(i, k), tag) is the first correction function
  • demo(distance(i, k), up) is the second correction function
  • tag is the tag attribute
  • up is the author attribute.
  • the first correction function and the second correction function are decay functions whose value range is [0, 1], including a linear function or a quadratic function.
  • the correction formula includes:
  • Ttag is the decay constant preset according to the tag attribute
  • Tup is the decay constant preset according to the author attribute.
  • the correction module 450 is further configured to: obtain the corresponding target user attribute from the user terminal to be recommended; according to the n and the sorting distance between each target video and the first video, Combining the attributes of the target user and/or the video type of the first video, the initial recommendation score of the first video is revised to obtain a revised recommendation score corresponding to the target user.
  • FIG. 6 schematically shows a schematic diagram of a hardware architecture of a computer device 1 suitable for implementing a video content recommendation method according to Embodiment 3 of the present application.
  • the computer device 1 is a device that can automatically perform numerical calculation and/or information processing according to pre-set or stored instructions.
  • it may be a rack server, blade server, tower server or cabinet server (including an independent server, or a server cluster composed of multiple servers) with a gateway function.
  • the computer device 1 at least includes but is not limited to: a memory 510 , a processor 520 , and a network interface 530 that can communicate with each other through a system bus. in:
  • the memory 510 includes at least one type of computer-readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (eg, SD or DX memory, etc.), random access memory (RAM), static random access memory, etc. (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk, optical disk, etc.
  • the memory 510 may be an internal storage module of the computer device 1 , such as a hard disk or a memory of the computer device 1 .
  • the memory 510 may also be an external storage device of the computer device 1, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC for short), a secure digital (Secure) Digital, referred to as SD) card, flash card (Flash Card) and so on.
  • the memory 510 may also include both an internal storage module of the computer device 1 and an external storage device thereof.
  • the memory 510 is generally used to store the operating system and various types of application software installed in the computer device 1 , such as program codes of a video content recommendation method, and the like.
  • the memory 510 may also be used to temporarily store various types of data that have been output or will be output.
  • the processor 520 may be a central processing unit (Central Processing Unit, CPU for short), a controller, a microcontroller, a microprocessor, or other data processing chips.
  • the processor 520 is generally used to control the overall operation of the computer device 1 , for example, to perform control and processing related to data interaction or communication with the computer device 1 .
  • the processor 520 is configured to run program codes or process data stored in the memory 510 .
  • the network interface 530 may comprise a wireless network interface or a wired network interface, and the network interface 530 is typically used to establish a communication link between the computer device 1 and other computer devices.
  • the network interface 530 is used to connect the computer device 1 with an external terminal through a network, and to establish a data transmission channel and a communication link between the computer device 1 and the external terminal, and so on.
  • the network can be Intranet, Internet, Global System of Mobile communication (GSM for short), Wideband Code Division Multiple Access (WCDMA for short), 4G network , 5G network, Bluetooth (Bluetooth), Wi-Fi and other wireless or wired networks.
  • FIG. 6 only shows a computer device having components 510-530, but it should be understood that implementation of all of the illustrated components is not required, and that more or fewer components may be implemented instead.
  • the program code of the video content recommendation method stored in the memory 510, or the program code of the video content recommendation method may also be divided into one or more program modules, which are executed by one or more processors (this The embodiments are executed by the processor 520) to complete the embodiments of the present application.
  • This embodiment also provides a computer-readable storage medium, where computer-readable instructions are stored on the computer-readable storage medium, and when the computer-readable instructions are executed by a processor, the following steps are implemented:
  • the computer-readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read-Only Memory (PROM), magnetic memory, magnetic disk, optical disk, etc.
  • the computer-readable storage medium may be an internal storage unit of a computer device, such as a hard disk or memory of the computer device.
  • the computer-readable storage medium may also be an external storage device of a computer device, such as a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC for short), a secure digital ( Secure Digital, referred to as SD) card, flash memory card (Flash Card) and so on.
  • the computer-readable storage medium may also include both an internal storage unit of a computer device and an external storage device thereof.
  • the computer-readable storage medium is generally used to store the operating system and various application software installed on the computer device, for example, the program code of the video content recommendation method in the embodiment.
  • the computer-readable storage medium can also be used to temporarily store various types of data that have been output or will be output.
  • each module or each step of the above-mentioned embodiments of the present application can be implemented by a general-purpose computing device, and they can be concentrated on a single computing device, or distributed in multiple computing devices. network, they can optionally be implemented with program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, can be different from the The illustrated or described steps are performed in order, either by fabricating them separately into individual integrated circuit modules, or by fabricating multiple modules or steps of them into a single integrated circuit module. As such, the embodiments of the present application are not limited to any specific combination of hardware and software.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)

Abstract

本申请公开了一种视频内容推荐方法、装置及计算机设备，该方法包括：将待选视频集的待选视频作为第一视频预添加到推荐视频序列，然后截取包含所述第一视频的视频序列区段，接着根据第一视频在所述视频序列区段中的位置序号以及具有第一视频的目标属性的其他视频在所述视频序列区段中的位置序号对第一视频的初始推荐评分进行修正，得到修正推荐评分，最后根据修正推荐评分选择最终的待选视频添加到所述推荐视频序列。本申请还提供一种计算机可读存储介质。本申请实现根据相似视频积累所有同类视频的负向影响力来达到多样性控制，因此，提高了推荐内容的多样性，也提高了推荐方式的灵活性。

Description

视频内容推荐方法、装置及计算机设备
本申请要求于2020年07月06日提交中国专利局、申请号为202010645506.3、发明名称为“视频内容推荐方法、装置及计算机设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及互联网技术领域，尤其涉及一种视频内容推荐方法、装置及计算机设备。
背景技术
随着互联网技术的发展，每天有海量的视频数据上载到视频平台，然后用户则可以基于该视频平台选择并播放对应的视频内容。而视频平台一般也会提供内容推荐功能，以提高视频平台的用户粘度。然而，发明人意识到，现有的大部分视频平台，其视频推荐系统一般都是根据用户个性化兴趣从视频库中检索出一定量的待选集，然后对这些待选集进行打分排序，再推荐到对应的用户端。用户对于某一类视频的点击次数越多或观看时长越长，获得该类推荐视频越多，而用户对于该类推荐视频的点击又会进一步强化对该类视频的推荐。
发明内容
本申请提出一种视频内容推荐方法、装置及计算机设备，能够解决现有技术中在视频内容推荐过程中推荐视频缺乏多样性，推荐灵活性较差的问题。
首先，为实现上述目的，本申请提供一种视频内容推荐方法，所述方法包括：
依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；获取所述第一视频的目标属性和初始推荐评分；在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；将所述第一视频放回所述待选视频集；以及从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
在一个例子中，所述获取所述第一视频的目标属性包括：当所述第一视频的目标属性 为空时，对所述第一视频标记上预设的视频属性，然后将所述视频属性作为所述第一视频的目标属性。
在一个例子中，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：当所述n大于或等于预设的阈值N时，将所述第一视频的修正推荐评分置为0。
在一个例子中，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：当所述n小于预设的阈值N时，获取所述目标视频在所述视频序列区段中的位置序号i以及所述第一视频在所述视频序列区段中的位置序号k；根据所述i、所述k以及预设的修正公式对所述初始推荐评分进行修正。
在一个例子中，所述目标属性包括标签属性和作者属性，所述修正公式包括：
Figure PCTCN2021101626-appb-000001
其中，scorek为所述第一视频的初始推荐评分，score’k为所述第一视频的修正推荐评分，count(tagsk)为所述第一视频的标签属性的个数，count(tagsi)为所述目标视频的标签属性的个数，demote(distance(i，k)，tag)为第一修正函数，demote(distance(i，k)，up)为第二修正函数，tag为标签属性，up为作者属性。
在一个例子中，所述第一修正函数和所述第二修正函数为值域为[0，1]的衰减函数，包括一次函数或二次函数。
在一个例子中，当所述第一修正函数和所述第二修正函数均为半衰期函数时，所述修正公式包括：
Figure PCTCN2021101626-appb-000002
其中，Ttag为根据标签属性预先设置的衰减常数，Tup为根据作者属性预先设置的衰减常数。
在一个例子中，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：从待推荐的用户终端获取对应的目标用户属性；根据所述n以及每一所述目标视频与所述第一视频的排序距离，结合所述目标用户属性和/或所述第一视频的视频类型，对所述第一视频的所述初始推荐评分进行修正，得到对应于所述目标用户的修正推荐评分。
此外，为实现上述目的，本申请还提供一种视频内容推荐装置，所述装置包括：
预添加模块，用于依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；截取模块，用于在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；获取模块，用于获取所述第一视频的目标属性和初始推荐评分；查找模块，用于在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；修正模块，用于根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；选择模块，用于将所述第一视频放回所述待选视频集；以及从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
进一步地，本申请还提出一种计算机设备，所述计算机设备包括存储器、处理器，所述存储器上存储有可在所述处理器上运行的计算机可读指令，所述计算机可读指令被所述处理器执行时实现以下步骤：
依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；获取所述第一视频的目标属性和初始推荐评分；在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；将所述第一视频放回所述待选视频集；以及从所述待选视频集中 选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
进一步地，为实现上述目的，本申请还提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令可被至少一个处理器执行，以使所述至少一个处理器执行以下步骤：
依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；获取所述第一视频的目标属性和初始推荐评分；在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；将所述第一视频放回所述待选视频集；以及从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
本申请所提出的视频内容推荐方法、装置、计算机设备及计算机可读存储介质，能够将待选视频集的待选视频作为第一视频预添加到推荐视频序列，然后截取包含所述第一视频的视频序列区段，接着根据第一视频在所述视频序列区段中的位置序号以及具有第一视频的目标属性的其他视频在所述视频序列区段中的位置序号对第一视频的初始推荐评分进行修正，得到修正推荐评分，最后根据修正推荐评分选择最终的待选视频添加到所述推荐视频序列。从而实现根据相似视频积累所有同类视频的负向影响力来达到多样性控制，因此，提高了推荐内容的多样性，也提高了推荐方式的灵活性。
附图说明
图1是本申请一实施例的应用环境示意图；
图2是本申请视频内容推荐方法一具体实施例的流程示意图；
图3是图2中步骤S208的一具体实施例的流程示意图；
图4是本申请一具体实施例的视频内容推荐方法的效果图；
图5是本申请视频内容推荐装置一实施例的程序模块示意图；
图6是本申请计算机设备一可选的硬件架构的示意图。
具体实施方式
申请人发现，现有技术中对于视频推荐的方式容易导致相同类型的视频的连续或高频率出现，从而降低用户的观看意愿，影响用户的使用体验。换句话说，现有的视频推荐系统很难保证推荐视频的多样性，推荐灵活性欠缺。
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
需要说明的是，在本申请中涉及“第一”、“第二”等的描述仅用于描述目的，而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。另外，各个实施例之间的技术方案可以相互结合，但是必须是以本领域普通技术人员能够实现为基础，当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在，也不在本申请要求的保护范围之内。
图1是本申请一实施例的应用环境示意图。参阅图1所示，所述计算机设备1与用户终端连接。所述计算机设备1能够获取到所述用户终端上的用户信息和用户行为记录，用户信息包括用户性别、职业、年龄以及地域等信息；用户行为记录包括用户观看记录、收藏记录、点击记录、点赞记录等其他互动记录。所述计算机设备1根据用户终端的用户信息和用户行为记录分析出该用户终端对应的用户的兴趣点，例如，通过统计用户观看记录，查找出用户对于某些视频类型的视频观看或收藏或点击或点赞次数较多，从而分析出该用户对该视频类型的视频比较偏好。
接着，所述计算机设备1根据所述用户的兴趣点从视频资源服务器或者数据库(图1未示出)中筛选出预设数量的视频作为待选视频集。当然，所述计算机设备1除了将与所述用户的兴趣点相符合的视频添加到所述待选视频集，还会选择其他热门视频，比如根据视频点赞率/关注率/互动率/点击率，或者热播评分等筛选出热门视频添加到所述待选视频集，以用于后续推荐给所述用户对应的用户终端。
在本实施例中，所述用户终端可作为手机、平板、便携设备、PC机或者其他具有显示功能的电子设备等；所述计算机设备1则可作为PC机、数据服务器等。
实施例一
图2是本申请视频内容推荐方法一实施例的流程示意图。可以理解，本方法实施例中的流程图不用于对执行步骤的顺序进行限定。下面以计算机设备1为执行主体进行示例性描述。
如图2所示，所述视频内容推荐方法可以包括步骤S200～S206。
步骤S200，依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的 最后序位。
步骤S202，在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段。
步骤S204，获取所述第一视频的目标属性和初始推荐评分。
具体的，所述计算机设备1根据用户的兴趣点筛选出预设数量的视频，同时选择出一定数量热门视频一同添加到待选视频集。接着，所述计算机设备1依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位。在本实施例中，预添加只是暂时性添加，所述计算机设备1通过依次将所述待选视频集中的待选视频预添加到所述推荐视频序列，从而可以依次对预添加到所述推荐视频序列中的待选视频进行后续的推荐评分的修正。
所述计算机设备1将所述待选视频集中的待选视频作为所述第一视频预添加到所述推荐视频序列之后，则进一步自所述第一视频起，向所述推荐视频序列的排序反方向截取预设数量视频的视频序列区段。
截取到所述视频序列区段之后，接着，所述计算机设备1获取所述第一视频的目标属性和初始推荐评分。在本实施例中，所述目标属性包括标签属性和作者属性，具体地，例如，用户在将视频上传到视频数据服务器时，一般会对该视频打上标签属性，比如，动漫标签，搞笑视频标签，娱乐综艺标签等等；然后上传到视频数据服务器的视频也会自动携带上传该视频的用户的信息，也就是作者属性。
当然，所述待选视频集中的视频还包括一个初始推荐评分，所述初始推荐评分可以理解为当所述计算机设备1将所述视频添加到所述待选视频集时，所述视频被评估出的推荐权重值。在本实施例中，所述推荐权重值可以是根据所述视频的点赞率/关注率/互动率/点击率等计算出来的，或者由观看用户评价给出的，或者由人工将视频播放情况以及点评情况综合评测出来。
由于所述计算机设备1从视频数据服务器筛选出一定数量的视频作为待选视频集，因此，所述计算机设备1可以获取到所述第一视频的目标属性；以及，获取所述第一视频的初始推荐评分。
在一个例子中，所述目标属性包括根据所述视频的相关信息所提取出的任意属性，所述相关信息包括但不限于所述视频的分类、简介、作者信息、上传者信息、评论信息、弹幕信息等。在一个例子中，可以通过对上述相关信息进行分析，并提取出上述信息中的关键词等方式，确定所述目标属性。
在一具体实施例中，所述目标属性包括但不限于所述目标视频的标签数据和/或作者属 性。在此，所述计算机设备1可以根据所述第一视频的视频分区、标题、简介、作者信息等，确定所述第一视频的标签属性；所述计算机设备1可以根据所述第一视频的作者信息，如作者的分区、粉丝数、历史上传视频的分类、历史上传视频的关键词、性别、年龄、爱好等，确定所述作者属性。
在一具体实施例中，当所述计算机设备获取所述第一视频的目标属性时，所述第一视频的目标属性为空，那么，所述计算机设备1则会对所述第一视频标记上预设的视频属性，然后将所述视频属性作为所述第一视频的目标属性。例如，当所述第一视频，由于用户上传该视频时未打上标签，那么所述计算机设备1在则不能获取到所述第一视频的标签属性，或者获取到的标签属性为无效标签，例如，不属于预设的规范标签，或者乱码显示。那么，所述计算机设备1则会对所述第一视频打上预设的标签属性。
步骤S206，在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n。
步骤S208，根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分。
具体的，所述计算机设备1在获取到所述第一视频的目标属性之后，则进一步获取所述视频序列区段中其他视频的属性值，然后找出具有所述目标属性的视频作为目标视频，并统计所有的目标视频的数量n。
在本实施例中，所述计算机设备1预设设置视频推荐的规则，比如，为了缓解用户观看视频的疲劳度，在推荐视频序列中的一个预设区段内，不能出现超过一定数量的相同的目标属性。因此，所述计算机设备1在统计出所述视频序列区段中所述目标视频的数量n之后，则进一步判断所述n是否大于或等于预设的阈值N。
当所述n大于或等于预设的阈值N时，所述计算机设备1则将所述第一视频的修正推荐评分置为0。例如，在所述视频序列区段中，具有相同标签属性的视频不能超过一半，若超过一半，比如10个超过5个视频都是带有动漫标签，那么，所述计算机设备1则会将预添加的带有动漫标签的所述第一视频的推荐评分置为0分。通过这种方式，可以对于连续出现的相同目标属性的视频进行限制，从而减少相同类型的视频连续或高频率出现。
参阅图3所示，当所述n小于预设的阈值N时，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：
步骤S300，获取所述目标视频在所述视频序列区段中的位置序号i以及所述第一视频在所述视频序列区段中的位置序号k。
步骤S302，根据所述i、所述k以及预设的修正公式对所述初始推荐评分进行修正。
在本实施例中，所述目标属性包括标签属性和作者属性，所述修正公式包括：
Figure PCTCN2021101626-appb-000003
其中，scorek为所述第一视频的初始推荐评分，score’k为所述第一视频的修正推荐评分，count(tagsk)为所述第一视频的标签属性的个数，count(tagsi)为所述目标视频的标签属性的个数，demote(distance(i，k)，tag)为第一修正函数，demote(distance(i，k)，up)为第二修正函数，tag为标签属性，up为作者属性。其中，所述第一修正函数和所述第二修正函数为值域为[0，1]的衰减函数，包括一次函数或二次函数。具体的，所述第一修正函数用于根据相同标签属性的目标视频与所述第一视频的位置序号之间的差值对所述第一视频的初始推荐评分进行修正；所述第二修正函数用于根据相同作者属性的目标视频与所述第一视频的位置序号之间的差值对所述第一视频的初始推荐评分进行修正。
在一具体实施例中，当所述第一修正函数和所述第二修正函数均为半衰期函数时，所述修正公式包括：
Figure PCTCN2021101626-appb-000004
其中，Ttag为根据标签属性预先设置的衰减常数，Tup为根据作者属性预先设置的衰减常数。在一具体实施例中，所述Ttag和所述Tup是通过离线评估或者上线实验得出的一个衰减参数，具体地：根据待选视频集的中视频的初始推荐评分，用户信息、用户行为记录以及所述待选视频集中的视频信息，比如所有视频包括的多种标签属性的分布情况，连续出现相同标签/视频作者的视频的概率，同类视频间隔为1/2/3/x的概率，连续5/6/x个视 频中出现同类视频的概率，每个用户观看到的视频分类的占比变化，以及选中视频的历史互动率指标等信息离线设置出所述Ttag和所述Tup；或者通过预设的在线推荐引擎创造一层多样性控制的实验，根据实验接入流量，测试在线推荐指标的优劣以及留存数据评价各组降权参数的实验，从而选择出最优的所述Ttag和所述Tup。
在另一具体实施例中，所述计算机设备1还可以从待推荐的用户终端获取对应的目标用户属性，然后根据所述n以及每一所述目标视频与所述第一视频的排序距离，结合所述目标用户属性和/或所述第一视频的视频类型，对所述第一视频的所述初始推荐评分进行修正，得到对应于所述目标用户的修正推荐评分。
具体的，在一种实施例中，所述目标用户属性以及所述第一视频的视频类型可以分别对应不同的修正参数，以对所述初始推荐评分进行修正，例如，对于不同的目标用户属性以及不同的视频类型，可以直接对视频的初始推荐评分进行加权或降权；或者，在另一种实施例中，由于每个用户对于连续观看相同类型的视频内容的接受程度不一样，因此，可以先获取待推荐的目标用户的目标用户属性，所述目标用户属性包括用户对于不同类型视频内容的衰减常数Ttag和Tup，因此，所述计算机设备1可以根据所述第一视频的视频类型，从获取到的所述目标用户属性中选择对应的Ttag和Tup，然后再结合到以上描述的修正推荐评分计算方法中，从而计算出对应于所述目标用户的修正推荐评分。
步骤S210，将所述第一视频放回所述待选视频集。
步骤S212，从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
在得到所述第一视频的修正推荐评分之后，所述计算机设备1则将所述第一视频放回至所述待选视频集中，当所述待选视频集中的所有视频的初始推荐评分均通过以上方式进行修正之后，所述计算机设备1则会从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
所述计算机设备1将待选视频集中的视频预添加到推荐视频序列，然后在推荐视频序列中建立评估生效范围/滑动窗口(即所述视频序列区段)，然后根据所述评估生效范围/滑动窗口中其他具有相同标签属性或作者属性的视频对预添加的视频的初始推荐评分进行修正。其中，可以引入生效范围/滑动窗口，比如与待选位置距离很远的同类视频，可以直接忽略掉，同时可以节省在线计算耗时。
请参阅图4，是本申请一示例性例的视频内容推荐方法的效果图。
在本实施例中，所述计算机设备1从待选视频集中选择第一个视频作为第一视频预添加推荐视频序列中的最后序号，也就是序号4的位置；然后将包含序号4的4个视频截取 为视频序列区段；获取第一视频的目标属性，即tag：0以及初始推荐评分score，然后查找出所述视频序列区段中具有tag：0的视频，也就是所述视频序列区段中的第1序号的视频；接着，所述计算机设备1根据以下公式：
Figure PCTCN2021101626-appb-000005
计算出所述第一视频的修正推荐评分score’，然后将所述第一视频放回至待选视频集。最后，所述计算机设备1根据这种方式依次对待选视频集中的视频的初始推荐评分进行修正后，再选择最高修正推荐评分的视频添加到推荐视频序列的序号4的位置。
本申请所述的计算机设备1能够将待选视频集的待选视频作为第一视频预添加到推荐视频序列，然后截取包含所述第一视频的视频序列区段，接着根据第一视频在所述视频序列区段中的位置序号以及具有第一视频的目标属性的其他视频在所述视频序列区段中的位置序号对第一视频的初始推荐评分进行修正，得到修正推荐评分，最后根据修正推荐评分选择最终的待选视频添加到所述推荐视频序列。从而实现根据相似视频积累所有同类视频的负向影响力来达到多样性控制，因此，提高了推荐内容的多样性，也提高了推荐方式的灵活性。
对于现有技术中对于视频内容推荐的方式，例如，根据简单的多样性限制规则，这个规则就是限制同类视频内容的推荐数量，比如，在视频内容推荐过程中，推荐给用户的5个视频中，同类视频不能超过2个。因此，将这个规则应用于所有用户，比较难平衡满足限制规则和推荐指标；而且这种规则中对不同类型视频的不同处理基于人的经验判断，而且无法进一步改进和扩展。再例如，根据MMR(Maximal Marginal Relevance，最大边界相关法)方式进行视频内容推荐，其中，MMR只选取相似度最大的视频来降权，返回的结果是同类视频的降权，这种方式没有考虑到在一个视频序列中的不同视频类型的视频降权，比如在我们推荐场景下的经验判断，连续出同类视频比间隔出同类视频体验更差；待选时只取已选同类视频中与其最大相关性的视频进行负向影响力，即降权，而没有多个同类的叠加。
与简单规则的方式相比，本申请可以支持更多的个性化推荐，对推荐出的视频分类占比有更好的控制能力，在用户消费角度，可以得到更好的用户满意度；与MMR方式相比，本申请可以支持多个已选中的同类视频对备选的负向影响力可以积累，所以如果一些原因已造成多个同类视频较高频率出现，后续同类视频出现的可能性更快速度下降；支持更多 个性化推荐能力，对不同的用户和不同类型的视频采取不同的处理。综上可知，本申请能够实现：
1.不同的标签/视频作者可以对后续同类视频施加不同程度的负向影响力，比如可以对壁纸视频的互斥性的设定更强，对搞笑的互斥性设定稍弱，负向影响力基于指标数据和运营方向的调整，可以有更好的表现力和解释性。
2.多个已选中的同类视频对备选的负向影响力可以积累，所以如果一些原因已造成多个同类视频较高频率出现，后续同类视频出现的可能性更快速度下降。
3.容易扩展支持针对不用用户对同类型视频的喜好程度(和耐看程度)作出更多个性化的推荐。
也就是说，本申请能综合考虑整个推荐的视频序列，待选视频与所有同类视频的相似度以及在序列中的位置距离，保证推荐系统的指标效果(点赞率/关注率/整体互动率等)，并且推荐在线系统中能以较少的资源消耗和响应时间完成整个的推荐计算。
实施例二
图5示意性示出了根据本申请实施例二的视频内容推荐装置的框图，该视频内容推荐装置可以被分割成一个或多个程序模块，一个或者多个程序模块被存储于存储介质中，并由一个或多个处理器所执行，以完成本申请实施例。本申请实施例所称的程序模块是指能够完成特定功能的一系列计算机可读指令段，以下描述将具体介绍本实施例中各程序模块的功能。
如图5所示，该视频内容推荐装置400可以包括预添加模块410、截取模块420、获取模块430、查找模块440、修正模块450和选择模块460，其中：
预添加模块410，用于依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位.
截取模块420，用于在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段。
获取模块430，用于获取所述第一视频的目标属性和初始推荐评分。
查找模块440，用于在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n。
修正模块450，用于根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分。
选择模块460，用于将所述第一视频放回所述待选视频集；以及从所述待选视频集中 选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
在示例性的实施例中，修正模块450，还用于：当所述n大于或等于预设的阈值N时，将所述第一视频的修正推荐评分置为0。当所述n小于预设的阈值N时，获取所述目标视频在所述视频序列区段中的位置序号i以及所述第一视频在所述视频序列区段中的位置序号k；根据所述i、所述k以及预设的修正公式对所述初始推荐评分进行修正。其中，所述目标属性包括标签属性和作者属性，所述修正公式包括：
Figure PCTCN2021101626-appb-000006
其中，scorek为所述第一视频的初始推荐评分，score’k为所述第一视频的修正推荐评分，count(tagsk)为所述第一视频的标签属性的个数，count(tagsi)为所述目标视频的标签属性的个数，demote(distance(i，k)，tag)为第一修正函数，demote(distance(i，k)，up)为第二修正函数，tag为标签属性，up为作者属性。
在一示例性例子中，所述第一修正函数和所述第二修正函数为值域为[0，1]的衰减函数，包括一次函数或二次函数。当所述第一修正函数和所述第二修正函数均为半衰期函数时，所述修正公式包括：
Figure PCTCN2021101626-appb-000007
其中，Ttag为根据标签属性预先设置的衰减常数，Tup为根据作者属性预先设置的衰减常数。
在示例性的实施例中，修正模块450，还用于：从待推荐的用户终端获取对应的目标用户属性；根据所述n以及每一所述目标视频与所述第一视频的排序距离，结合所述目标 用户属性和/或所述第一视频的视频类型，对所述第一视频的所述初始推荐评分进行修正，得到对应于所述目标用户的修正推荐评分。
实施例三
图6示意性示出了根据本申请实施例三的适于实现视频内容推荐方法的计算机设备1的硬件架构示意图。本实施例中，计算机设备1是一种能够按照事先设定或者存储的指令，自动进行数值计算和/或信息处理的设备。例如，可以是具有网关功能的机架式服务器、刀片式服务器、塔式服务器或机柜式服务器(包括独立的服务器，或者多个服务器所组成的服务器集群)等。如图14所示，计算机设备1至少包括但不限于：可通过系统总线相互通信链接存储器510、处理器520、网络接口530。其中：
存储器510至少包括一种类型的计算机可读存储介质，可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中，存储器510可以是计算机设备1的内部存储模块，例如该计算机设备1的硬盘或内存。在另一些实施例中，存储器510也可以是计算机设备1的外部存储设备，例如该计算机设备1上配备的插接式硬盘，智能存储卡(Smart Media Card，简称为SMC)，安全数字(Secure Digital，简称为SD)卡，闪存卡(Flash Card)等。当然，存储器510还可以既包括计算机设备1的内部存储模块也包括其外部存储设备。本实施例中，存储器510通常用于存储安装于计算机设备1的操作系统和各类应用软件，例如视频内容推荐方法的程序代码等。此外，存储器510还可以用于暂时地存储已经输出或者将要输出的各类数据。
处理器520在一些实施例中可以是中央处理器(Central Processing Unit，简称为CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器520通常用于控制计算机设备1的总体操作，例如执行与计算机设备1进行数据交互或者通信相关的控制和处理等。本实施例中，处理器520用于运行存储器510中存储的程序代码或者处理数据。
网络接口530可包括无线网络接口或有线网络接口，该网络接口530通常用于在计算机设备1与其他计算机设备之间建立通信链接。例如，网络接口530用于通过网络将计算机设备1与外部终端相连，在计算机设备1与外部终端之间的建立数据传输通道和通信链接等。网络可以是企业内部网(Intranet)、互联网(Internet)、全球移动通讯系统(Global System of Mobile communication，简称为GSM)、宽带码分多址(Wideband Code Division Multiple Access，简称为WCDMA)、4G网络、5G网络、蓝牙(Bluetooth)、Wi-Fi等无线或有线网 络。
需要指出的是，图6仅示出了具有部件510-530的计算机设备，但是应理解的是，并不要求实施所有示出的部件，可以替代的实施更多或者更少的部件。
在本实施例中，存储于存储器510中的视频内容推荐方法的程序代码，或者视频内容推荐方法的程序代码还可以被分割为一个或者多个程序模块，并由一个或多个处理器(本实施例为处理器520)所执行，以完成本申请实施例。
实施例四
本实施例还提供一种计算机可读存储介质，计算机可读存储介质其上存储有计算机可读指令，计算机可读指令被处理器执行时实现以下步骤：
依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；获取所述第一视频的目标属性和初始推荐评分；在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；将所述第一视频放回所述待选视频集；以及从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
本实施例中，计算机可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中，计算机可读存储介质可以是计算机设备的内部存储单元，例如该计算机设备的硬盘或内存。在另一些实施例中，计算机可读存储介质也可以是计算机设备的外部存储设备，例如该计算机设备上配备的插接式硬盘，智能存储卡(Smart Media Card，简称为SMC)，安全数字(Secure Digital，简称为SD)卡，闪存卡(Flash Card)等。当然，计算机可读存储介质还可以既包括计算机设备的内部存储单元也包括其外部存储设备。本实施例中，计算机可读存储介质通常用于存储安装于计算机设备的操作系统和各类应用软件，例如实施例中视频内容推荐方法的程序代码等。此外，计算机可读存储介质还可以用于暂时地存储已经输出或者将要输出的各类数据。
显然，本领域的技术人员应该明白，上述的本申请实施例的各模块或各步骤可以用通用的计算装置来实现，它们可以集中在单个的计算装置上，或者分布在多个计算装置所组 成的网络上，可选地，它们可以用计算装置可执行的程序代码来实现，从而，可以将它们存储在存储装置中由计算装置来执行，并且在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本申请实施例不限制于任何特定的硬件和软件结合。
以上仅为本申请实施例的优选实施例，并非因此限制本申请实施例的专利范围，凡是利用本申请实施例说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请实施例的专利保护范围内。

Claims (20)

  1. 一种视频内容推荐方法，所述方法包括：
    依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；
    在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；
    获取所述第一视频的目标属性和初始推荐评分；
    在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；
    根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；
    将所述第一视频放回所述待选视频集；以及
    从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
  2. 如权利要求1所述的视频内容推荐方法，所述获取所述第一视频的目标属性包括：
    当所述第一视频的目标属性为空时，对所述第一视频标记上预设的视频属性，然后将所述视频属性作为所述第一视频的目标属性。
  3. 如权利要求1或2所述的视频内容推荐方法，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：
    当所述n大于或等于预设的阈值N时，将所述第一视频的修正推荐评分置为0。
  4. 如权利要求1-3中任一项所述的视频内容推荐方法，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：
    当所述n小于预设的阈值N时，获取所述目标视频在所述视频序列区段中的位置序号i以及所述第一视频在所述视频序列区段中的位置序号k；
    根据所述i、所述k以及预设的修正公式对所述初始推荐评分进行修正。
  5. 如权利要求4所述的视频内容推荐方法，所述目标属性包括标签属性和作者属性，所述修正公式包括：
    Figure PCTCN2021101626-appb-100001
    其中，scorek为所述第一视频的初始推荐评分，score’k为所述第一视频的修正推荐评分，count(tagsk)为所述第一视频的标签属性的个数，count(tagsi)为所述目标视频的标签属性的个数，demote(distance(i，k)，tag)为第一修正函数，demote(distance(i，k)，up)为第二修正函数，tag为标签属性，up为作者属性。
  6. 如权利要求5所述的视频内容推荐方法，所述第一修正函数和所述第二修正函数为值域为[0，1]的衰减函数，包括一次函数或二次函数。
  7. 如权利要求5所述的视频内容推荐方法，当所述第一修正函数和所述第二修正函数均为半衰期函数时，所述修正公式包括：
    Figure PCTCN2021101626-appb-100002
    其中，Ttag为根据标签属性预先设置的衰减常数，Tup为根据作者属性预先设置的衰减常数。
  8. 如权利要求1-7中任一项所述的视频内容推荐方法，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：
    从待推荐的用户终端获取对应的目标用户属性；
    根据所述n以及每一所述目标视频与所述第一视频的排序距离，结合所述目标用户属性和/或所述第一视频的视频类型，对所述第一视频的所述初始推荐评分进行修正，得到对应于所述目标用户的修正推荐评分。
  9. 一种视频内容推荐装置，所述装置包括：
    预添加模块，用于依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；
    截取模块，用于在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；
    获取模块，用于获取所述第一视频的目标属性和初始推荐评分；
    查找模块，用于在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；
    修正模块，用于根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；
    选择模块，用于将所述第一视频放回所述待选视频集；以及从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
  10. 一种计算机设备，所述计算机设备包括存储器、处理器，所述存储器上存储有可在所述处理器上运行的计算机可读指令，所述计算机可读指令被所述处理器执行时实现以下步骤：
    依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；
    在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；
    获取所述第一视频的目标属性和初始推荐评分；
    在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；
    根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；
    将所述第一视频放回所述待选视频集；以及
    从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
  11. 如权利要求10所述的计算机设备，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：
    当所述n大于或等于预设的阈值N时，将所述第一视频的修正推荐评分置为0；和/或
    当所述n小于预设的阈值N时，获取所述目标视频在所述视频序列区段中的位置序号i以及所述第一视频在所述视频序列区段中的位置序号k；
    根据所述i、所述k以及预设的修正公式对所述初始推荐评分进行修正；和/或
    从待推荐的用户终端获取对应的目标用户属性；
    根据所述n以及每一所述目标视频与所述第一视频的排序距离，结合所述目标用户属性和/或所述第一视频的视频类型，对所述第一视频的所述初始推荐评分进行修正，得到对应于所述目标用户的修正推荐评分。
  12. 如权利要求10或11所述的计算机设备，所述获取所述第一视频的目标属性包括：
    当所述第一视频的目标属性为空时，对所述第一视频标记上预设的视频属性，然后将所述视频属性作为所述第一视频的目标属性。
  13. 如权利要求10-12中任一项所述的计算机设备，所述目标属性包括标签属性和作者属性，所述修正公式包括：
    Figure PCTCN2021101626-appb-100003
    其中，scorek为所述第一视频的初始推荐评分，score’k为所述第一视频的修正推荐评分，count(tagsk)为所述第一视频的标签属性的个数，count(tagsi)为所述目标视频的标签属性的个数，demote(distance(i，k)，tag)为第一修正函数，demote(distance(i，k)，up)为第二修正函数，tag为标签属性，up为作者属性。
  14. 如权利要求13所述的计算机设备，所述第一修正函数和所述第二修正函数为值域为[0，1]的衰减函数，包括一次函数或二次函数；
    当所述第一修正函数和所述第二修正函数均为半衰期函数时，所述修正公式包括：
    Figure PCTCN2021101626-appb-100004
    其中，Ttag为根据标签属性预先设置的衰减常数，Tup为根据作者属性预先设置的衰减常数。
  15. 一种计算机可读存储介质，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令可被至少一个处理器执行，以使所述至少一个处理器执行以下步骤：
    依次从待选视频集中选取待选视频作为第一视频预添加到推荐视频序列的最后序位；
    在所述推荐视频序列中截取包含所述第一视频的预设数量视频的视频序列区段；
    获取所述第一视频的目标属性和初始推荐评分；
    在所述视频序列区段中查找出除去所述第一视频之外的具有所述目标属性的目标视频，并统计出所述目标视频的数量n；
    根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分；
    将所述第一视频放回所述待选视频集；以及
    从所述待选视频集中选择出修正推荐评分最高的视频添加到所述待推荐视频序列的最后序位。
  16. 如权利要求15所述的计算机可读存储介质，所述根据所述n以及每一所述目标视频与所述第一视频的排序距离对所述第一视频的所述初始推荐评分进行修正，得到修正推荐评分，包括：
    当所述n大于或等于预设的阈值N时，将所述第一视频的修正推荐评分置为0；和/或
    当所述n小于预设的阈值N时，获取所述目标视频在所述视频序列区段中的位置序号i以及所述第一视频在所述视频序列区段中的位置序号k；
    根据所述i、所述k以及预设的修正公式对所述初始推荐评分进行修正；和/或
    从待推荐的用户终端获取对应的目标用户属性；
    根据所述n以及每一所述目标视频与所述第一视频的排序距离，结合所述目标用户属性和/或所述第一视频的视频类型，对所述第一视频的所述初始推荐评分进行修正，得到对应于所述目标用户的修正推荐评分。
  17. 如权利要求15或16所述的计算机可读存储介质，所述获取所述第一视频的目标属性包括：
    当所述第一视频的目标属性为空时，对所述第一视频标记上预设的视频属性，然后将所述视频属性作为所述第一视频的目标属性。
  18. 如权利要求15-17中任一项所述的计算机可读存储介质，所述目标属性包括标签属性和作者属性，所述修正公式包括：
    Figure PCTCN2021101626-appb-100005
    其中，scorek为所述第一视频的初始推荐评分，score’k为所述第一视频的修正推荐评分，count(tagsk)为所述第一视频的标签属性的个数，count(tagsi)为所述目标视频的标签属性的个数，demote(distance(i，k)，tag)为第一修正函数，demote(distance(i，k)，up)为第二修正函数，tag为标签属性，up为作者属性。
  19. 如权利要求18所述的计算机可读存储介质，所述第一修正函数和所述第二修正函数为值域为[0，1]的衰减函数，包括一次函数或二次函数。
  20. 如权利要求19所述的计算机可读存储介质，当所述第一修正函数和所述第二修正函数均为半衰期函数时，所述修正公式包括：
    Figure PCTCN2021101626-appb-100006
    其中，Ttag为根据标签属性预先设置的衰减常数，Tup为根据作者属性预先设置的衰减常数。
PCT/CN2021/101626 2020-07-06 2021-06-22 视频内容推荐方法、装置及计算机设备 WO2022007626A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/014,339 US20230300417A1 (en) 2020-07-06 2021-06-22 Video content recommendation method and apparatus, and computer device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010645506.3 2020-07-06
CN202010645506.3A CN112423134B (zh) 2020-07-06 2020-07-06 视频内容推荐方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2022007626A1 true WO2022007626A1 (zh) 2022-01-13

Family

ID=74844218

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/101626 WO2022007626A1 (zh) 2020-07-06 2021-06-22 视频内容推荐方法、装置及计算机设备

Country Status (3)

Country Link
US (1) US20230300417A1 (zh)
CN (1) CN112423134B (zh)
WO (1) WO2022007626A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112423134B (zh) * 2020-07-06 2022-03-29 上海哔哩哔哩科技有限公司 视频内容推荐方法、装置、计算机设备及存储介质
CN112965776A (zh) * 2021-03-11 2021-06-15 广州酷狗计算机科技有限公司 屏保图片的显示方法、装置、终端及计算机可读存储介质
CN116738052A (zh) * 2021-07-09 2023-09-12 小红书科技有限公司 一种基于滑动谱分解确定内容多样性的方法和选定内容的方法

Citations (9)

* 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
US9152708B1 (en) * 2009-12-14 2015-10-06 Google Inc. Target-video specific co-watched video clusters
CN109522426A (zh) * 2018-12-05 2019-03-26 北京达佳互联信息技术有限公司 多媒体数据推荐方法、装置、设备及计算机可读存储介质
CN109710805A (zh) * 2018-12-13 2019-05-03 百度在线网络技术（北京）有限公司 基于兴趣簇的视频交互方法和装置
CN110096613A (zh) * 2019-04-12 2019-08-06 北京奇艺世纪科技有限公司 一种视频推荐方法、装置、电子设备及存储介质
US20190286673A1 (en) * 2016-11-21 2019-09-19 Comcast Cable Communications, Llc Content Recommendation System With Weighted Metadata Annotations
CN110730369A (zh) * 2019-10-15 2020-01-24 青岛聚看云科技有限公司 一种视频推荐方法及服务器
US10657144B2 (en) * 2016-08-18 2020-05-19 Google Llc Optimizing digital video distribution
CN112423134A (zh) * 2020-07-06 2021-02-26 上海哔哩哔哩科技有限公司 视频内容推荐方法、装置及计算机设备

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11070881B1 (en) * 2020-07-07 2021-07-20 Verizon Patent And Licensing Inc. Systems and methods for evaluating models that generate recommendations

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9152708B1 (en) * 2009-12-14 2015-10-06 Google Inc. Target-video specific co-watched video clusters
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
US10657144B2 (en) * 2016-08-18 2020-05-19 Google Llc Optimizing digital video distribution
US20190286673A1 (en) * 2016-11-21 2019-09-19 Comcast Cable Communications, Llc Content Recommendation System With Weighted Metadata Annotations
CN109522426A (zh) * 2018-12-05 2019-03-26 北京达佳互联信息技术有限公司 多媒体数据推荐方法、装置、设备及计算机可读存储介质
CN109710805A (zh) * 2018-12-13 2019-05-03 百度在线网络技术（北京）有限公司 基于兴趣簇的视频交互方法和装置
CN110096613A (zh) * 2019-04-12 2019-08-06 北京奇艺世纪科技有限公司 一种视频推荐方法、装置、电子设备及存储介质
CN110730369A (zh) * 2019-10-15 2020-01-24 青岛聚看云科技有限公司 一种视频推荐方法及服务器
CN112423134A (zh) * 2020-07-06 2021-02-26 上海哔哩哔哩科技有限公司 视频内容推荐方法、装置及计算机设备

Also Published As

Publication number Publication date
CN112423134B (zh) 2022-03-29
CN112423134A (zh) 2021-02-26
US20230300417A1 (en) 2023-09-21

Similar Documents

Publication Publication Date Title
US10405016B2 (en) Recommending media items based on take rate signals
WO2022007626A1 (zh) 视频内容推荐方法、装置及计算机设备
CN109819284B (zh) 一种短视频推荐方法、装置、计算机设备及存储介质
TWI636416B (zh) 內容個人化之多相排序方法和系統
WO2017181612A1 (zh) 个性化视频推荐方法及装置
CN106326391B (zh) 多媒体资源推荐方法及装置
CN107273508B (zh) 基于人工智能的信息处理方法和装置
US9552399B1 (en) Displaying information about distributed group discussions
US10387915B2 (en) Digital magazine recommendations by topic
EP3549032B1 (en) Determination of similarity between videos using shot duration correlation
WO2020238502A1 (zh) 物品推荐方法及装置、电子设备及存储介质
EP3356951B1 (en) Managing a database of patterns used to identify subsequences in logs
CN109168047B (zh) 视频推荐方法、装置、服务器及存储介质
US10346496B2 (en) Information category obtaining method and apparatus
CN108540508B (zh) 用于推送信息的方法、装置和设备
CN106897398B (zh) 一种视频展示方法及装置
EP3346396A1 (en) Multimedia resource quality assessment method and apparatus
CN110175264A (zh) 视频用户画像的构建方法、服务器及计算机可读存储介质
JP5228584B2 (ja) 興味情報特定システム、興味情報特定方法、および興味情報特定用プログラム
CN110895594A (zh) 一种页面展示的方法以及相关设备
CN111651666A (zh) 用户主题推荐方法、装置、计算机设备及存储介质
JP5813052B2 (ja) 情報処理装置、方法及びプログラム
CN109213933B (zh) 内容项推荐方法、装置、设备及存储介质
US20160267087A1 (en) Enhanced template curating
CN110971973A (zh) 一种视频推送方法、装置及电子设备

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: 21837506

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21837506

Country of ref document: EP

Kind code of ref document: A1