Video recommendation method, device, computer equipment and storage medium

Abstract

The embodiments of the present application disclose a video recommendation method, apparatus, computer equipment and storage medium; the embodiments of the present application can obtain a video interaction sequence of an object; perform intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, which includes intensity feature information corresponding to each video; perform multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, which includes at least one interest feature information corresponding to the object; fuse the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence and the interest feature information in the multi-interest feature sequence to obtain video interaction feature information for the object; perform video recommendation on the object based on the video interaction feature information, which can improve the accuracy of video recommendation.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a video recommendation method, apparatus, computer device, and storage medium. Background Art

[0002] With the development of computer technology, multimedia applications have become increasingly widespread, and the number of videos has increased dramatically. Watching videos has become a daily pastime for people. Since there are a wide variety of videos online, and every user has their own favorite videos, recommending suitable videos to users from this vast array of videos is an important and meaningful task. The inventors of this application have discovered through practical research on existing technologies that existing video recommendation methods suffer from low recommendation accuracy. Summary of the Invention

[0003] The embodiments of the present application provide a video recommendation method, apparatus, computer device, and storage medium, which can improve the accuracy of video recommendations.

[0004] The present invention provides a video recommendation method, including:

[0005] Acquire a video interaction sequence of an object, wherein the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information;

[0006] Performing intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video;

[0007] Performing multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object;

[0008] fusing the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multiple interest feature sequences to obtain video interaction feature information for the object;

[0009] Video recommendations are performed for the object based on the video interaction feature information.

[0010] Accordingly, an embodiment of the present application further provides a video recommendation device, including:

[0011] an acquiring unit, configured to acquire a video interaction sequence of an object, wherein the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information;

[0012] an intensity mapping unit, configured to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video;

[0013] a multi-interest mapping unit, configured to perform multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, wherein the multi-interest feature sequence includes at least one interest feature information corresponding to the object;

[0014] a fusion unit, configured to fuse the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multiple interest feature sequence to obtain video interaction feature information for the object;

[0015] A video recommendation unit is used to recommend videos for the object based on the video interaction feature information.

[0016] In one embodiment, the intensity mapping unit may include:

[0017] Mode acquisition subunit, used to obtain multiple intensity mapping modes;

[0018] an intensity mapping subunit, configured to perform intensity mapping of the attribute information and object interaction feature information corresponding to each video in multiple different dimensions according to the multiple intensity mapping modes, to obtain intensity feature sub-information of each video in multiple different dimensions;

[0019] The fusion subunit is used to fuse the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature information corresponding to each video.

[0020] In one embodiment, the intensity mapping subunit may include:

[0021] A first intensity mapping module is configured to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the time intensity mapping method to obtain intensity feature sub-information of each video in the time dimension;

[0022] A second intensity mapping module is configured to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the duration intensity mapping method to obtain intensity feature sub-information of each video in the duration dimension;

[0023] The third intensity mapping module is used to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the playback intensity mapping method to obtain intensity feature sub-information of each video in the playback dimension.

[0024] In one embodiment, the first intensity mapping module may include:

[0025] A first parsing submodule is configured to parse attribute information corresponding to the video according to the time-intensity mapping method to obtain video quantity threshold information, and parse the object interaction feature information to obtain position information of the video in the video interaction sequence;

[0026] a conversion submodule, configured to convert the position information of the video in the video interaction sequence into video reference degree information based on the video quantity threshold information according to the time intensity mapping mode;

[0027] The scaling submodule is configured to scale the video reference degree information according to the time intensity mapping method to obtain intensity feature sub-information of the video in the time dimension.

[0028] In one embodiment, the second intensity mapping module may include:

[0029] A second parsing submodule is configured to parse the attribute information corresponding to the video according to the duration-intensity mapping method to obtain duration threshold information, and parse the object interaction feature information corresponding to the video to obtain actual playback time information;

[0030] A first comparison submodule is configured to compare the actual playback duration information with the duration threshold information according to the duration intensity mapping method to obtain target duration information;

[0031] The normalization submodule is used to normalize the target duration information based on the duration threshold information according to the duration intensity mapping method to obtain the intensity feature sub-information of the video in the duration dimension.

[0032] In one embodiment, the third intensity mapping module may include:

[0033] A third parsing submodule is configured to parse the attribute information corresponding to the video according to the playback intensity mapping method to obtain playback progress threshold information and video duration information, and to parse the object interaction feature information corresponding to the video to obtain actual playback time information;

[0034] an arithmetic operation submodule, configured to perform an arithmetic operation on the video duration information of the video and the actual playback information according to the playback intensity mapping mode to obtain playback information after the operation;

[0035] The second comparison submodule is configured to compare the calculated playback information with the playback progress threshold information according to the playback intensity mapping method to obtain intensity feature sub-information of the video in the playback dimension.

[0036] In one embodiment, the first fusion subunit may include:

[0037] A first encoding module is used to encode the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature encoding information of each video in multiple different dimensions;

[0038] A first parameter determination module is used to determine the fusion parameter corresponding to the intensity feature coding information of each dimension of the video;

[0039] The first fusion module is used to fuse the intensity feature coding information of the video in multiple different dimensions based on the fusion parameters corresponding to the intensity feature coding information of each dimension to obtain intensity feature information corresponding to each video.

[0040] In one embodiment, the multi-interest mapping unit may include:

[0041] A parsing subunit, configured to parse the attribute information corresponding to each video in the video interaction sequence to obtain category information and tag information corresponding to each video;

[0042] The multi-interest mapping subunit is used to perform multi-interest mapping on the category information and tag information corresponding to each video to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object.

[0043] In one embodiment, the multi-interest mapping subunit may include:

[0044] A second encoding module is configured to encode the at least two category sub-information and the label information of each video to obtain category encoding feature information corresponding to the at least two category sub-information and label encoding feature information corresponding to the label information;

[0045] A second parameter determination module is used to determine the fusion parameters corresponding to the at least two category encoding feature information and the fusion parameters corresponding to the label encoding feature information;

[0046] The second fusion module is used to fuse the at least two category coding feature information and label coding feature information corresponding to each video based on the fusion parameters corresponding to the at least two category coding feature information of each video and the fusion parameters corresponding to the label coding feature information to obtain at least one interest feature information corresponding to the object.

[0047] In one embodiment, the second parameter determination module may include:

[0048] Get submodule, used to get object attribute information of an object;

[0049] A feature extraction submodule is used to extract features from the attribute information of the object to obtain object feature information;

[0050] a similarity information calculation submodule, configured to calculate category similarity information between the object feature information and the at least two category encoding feature information, and to calculate label similarity information between the object feature information and the label encoding feature information;

[0051] The parameter generation submodule is used to generate fusion parameters corresponding to the at least two category encoding feature information and fusion parameters corresponding to the label encoding feature information based on the category similarity information and the label similarity information.

[0052] In one embodiment, the video recommendation unit may include:

[0053] An information acquisition subunit, used to acquire object attribute information of an object;

[0054] a splicing subunit, configured to splice the video interaction feature information and the object attribute information of the object to obtain splicing feature information;

[0055] a multi-layer perception processing subunit, configured to perform multi-layer perception processing on the splicing feature information to obtain object feature representation information;

[0056] The video recommendation subunit is used to recommend videos for the object based on the object feature representation information.

[0057] The present application also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method provided in various optional embodiments of the above-mentioned aspect.

[0058] Correspondingly, an embodiment of the present application further provides a storage medium, which stores instructions. When the instructions are executed by a processor, any video recommendation method provided in the embodiment of the present application is implemented.

[0059] An embodiment of the present application can obtain a video interaction sequence of an object, where the video interaction sequence includes attribute information and object interaction feature information corresponding to at least one video; perform intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, where the video intensity feature sequence includes intensity feature information corresponding to each video, where the intensity feature information represents the interaction intensity between the object and each video; perform multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object; fuse the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multi-interest feature sequence to obtain video interaction feature information for the object; perform video recommendation for the object based on the video interaction feature information, thereby improving the accuracy of video recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0061] Figure 1 Schematic diagram of a video recommendation method according to an embodiment of the present invention;

[0062] Figure 2 Schematic diagram of the video recommendation method provided in the embodiment of the present application;

[0063] Figure 3 This is another scenario diagram of the video recommendation method provided in an embodiment of the present application;

[0064] Figure 4 This is a schematic diagram of the scenario of the title information provided in the embodiment of the present application;

[0065] Figure 5 This is another scenario diagram of the video recommendation method provided in an embodiment of the present application;

[0066] Figure 6 This is another flowchart of the video recommendation method provided in the embodiment of the present application;

[0067] Figure 7 Schematic diagram of the structure of the video recommendation device provided in an embodiment of the present application;

[0068] Figure 8 It is a structural diagram of the computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. However, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0070] The embodiments of the present application provide a video recommendation method. This video recommendation method can be performed by a video recommendation device, which can be integrated into a computer device. The computer device can include at least one of a terminal and a server. That is, the video recommendation method provided in the embodiments of the present application can be performed by a terminal, a server, or a terminal and a server that can communicate with each other.

[0071] Among them, terminals may include but are not limited to smartphones, tablets, laptops, personal computers (PCs), smart home appliances, wearable electronic devices, VR / AR devices, vehicle-mounted terminals, intelligent voice interaction devices, etc.

[0072] The server can be an intercommunication server or background server between multiple heterogeneous systems, an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and big data and artificial intelligence platforms, etc.

[0073] It should be noted that the embodiments of the present application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving, etc.

[0074] In one embodiment, if Figure 1The video recommendation device can be integrated into a computer device such as a terminal or a server to implement the video recommendation method proposed in the embodiment of the present application. Specifically, the server 11 can obtain a video interaction sequence of an object, the video interaction sequence including attribute information and object interaction feature information corresponding to at least one video; perform intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, the video intensity feature sequence including intensity feature information corresponding to each video, the intensity feature information representing the interaction intensity between the object and each video; perform multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, the multi-interest feature sequence including at least one interest feature information corresponding to the object; fuse the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multi-interest feature sequence to obtain video interaction feature information for the object; the server 11 performs video recommendation for the object based on the video interaction feature information. Then, the server 11 can send the video recommended to the object to the terminal 10, so that the object can watch the recommended video through the terminal 10.

[0075] The following are detailed descriptions of each embodiment. It should be noted that the description order of the following embodiments does not limit the preferred order of the embodiments.

[0076] The embodiments of the present application will be described from the perspective of a video recommendation device, which can be integrated into a computer device, which can be a server, a terminal, or other device.

[0077] like Figure 2 A video recommendation method is provided, and the specific process includes:

[0078] 101. Obtain a video interaction sequence of an object, where the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information.

[0079] The object's video interaction sequence is used to store information about videos on which the object has performed interactive operations. Interaction operations may include watching, liking, adding to favorites, sharing, and so on. For example, when the object watches a video, the video recommendation device may add the video's information to the video interaction sequence.

[0080] In one embodiment, the video interaction sequence may include attribute information corresponding to at least one video and object interaction feature information.

[0081] The attribute information of a video can be used to describe information that exists in the video itself and is not affected by the interaction of objects. For example, the attribute information of a video can include video category information, tag information, video duration information, etc.

[0082] The object interaction feature information of a video may include information generated during the object's interaction with the video. For example, the object interaction feature information of a video may include the actual playback time of the video viewed by the object, the position of the video in the video interaction sequence, and so on.

[0083] For example, suppose the video interaction sequence of the object is F = [f1, f2, ..., f i ,…,f n ]. Among them, f i It can represent the attribute information and object interaction feature information corresponding to at least one video in a video sequence.

[0084] In one embodiment, videos of objects performing interactive operations may be collected according to a preset number or a preset time, and information of the collected videos may be added to a video interaction sequence.

[0085] For example, assuming the preset number is 50, information of the 50 videos most recently watched by the subject may be collected and added to the video interaction sequence.

[0086] For another example, assuming that the preset time period may be one week, information about videos watched by the subject within the week may be collected and added to the video interaction sequence.

[0087] When adding a video to a video interaction sequence, the video can be added to the video interaction sequence in a preset order. For example, when constructing a video that collects interactive operations of a preset number of objects, the videos can be added to the video interaction sequence in the order in which the object watches the videos. For example, the object watched video A, video B, and video C. Among them, video A is watched first, video B is watched second, and video C is watched third. In this case, video A can be added to the video interaction sequence first, then video B can be added to the video interaction sequence, and finally video C can be added to the video interaction sequence.

[0088] 102. Perform intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence. The video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video.

[0089] In one embodiment, the embodiment of the present application can perform intensity mapping on the attribute information of the video and the object interaction feature information to obtain deeper video interaction feature information of the object through the attribute information of the video and the object interaction feature information. Then, video recommendations can be made for the object based on the video interaction feature information, thereby improving the accuracy of video recommendations.

[0090] In one embodiment, intensity mapping may be performed on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence.

[0091] The video strength feature sequence includes strength feature information corresponding to each video. The strength feature information represents the strength of interaction between the object and each video. For example, the strength feature information can be an indicator used to evaluate the strength of interaction between the object and each video. A strong interaction between the object and the video indicates that the user is more interested in the video, meaning that the video is more aligned with the user's viewing preferences. A weak interaction between the object and the video indicates that the user is less interested in the video.

[0092] In one embodiment, the intensity feature information may be represented in a variety of different forms.

[0093] For example, the intensity feature information may be a numerical value. When the interaction strength between the object and the video is strong, the numerical value of the intensity feature information may be large. On the other hand, when the interaction strength between the object and the video is weak, the numerical value of the intensity feature information may be small. For another example, the intensity feature information may be a vector, and so on.

[0094] Intensity mapping of the attribute information and object interaction feature information corresponding to each video in the video interaction sequence can refer to processing the attribute information and object feature information together to obtain an indicator for evaluating the interaction strength between the object and each video. The intensity mapping process of the attribute information and object interaction feature information corresponding to each video in the video interaction sequence can also be viewed as a process of information mining the attribute information and object interaction feature information corresponding to each video. By mining the attribute information and object interaction feature information corresponding to each video, the user's preferences and interests in watching videos can be determined, thereby improving the accuracy of video recommendations for the user.

[0095] In one embodiment, to further improve the accuracy of video recommendations, the embodiment of the present application may perform intensity mapping on the attribute information and object interaction feature information of the video using a variety of different intensity mapping methods. Specifically, the step of "performing intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video" may include:

[0096] Get multiple intensity mapping methods;

[0097] According to multiple intensity mapping methods, the attribute information and object interaction feature information corresponding to each video are subjected to intensity mapping in multiple different dimensions to obtain the intensity feature sub-information of each video in multiple different dimensions;

[0098] The intensity feature sub-information of each video in multiple different dimensions is fused to obtain the intensity feature information corresponding to each video.

[0099] The intensity mapping method may be a logic that needs to be followed when mapping the attribute information and object interaction feature information corresponding to the video into the intensity feature sub-information. For example, the intensity mapping method may be a predefined formula, etc.

[0100] In one embodiment, the intensity feature sub-information is also used to characterize the interaction strength between an object and a video. Different intensity feature sub-information can characterize the interaction strength between an object and a video from different evaluation dimensions. For example, intensity feature sub-information in different dimensions can include intensity feature sub-information in the time dimension, intensity feature sub-information in the duration dimension, and intensity feature sub-information in the playback dimension, etc.

[0101] In one embodiment, the intensity feature information may be an overall indicator that integrates intensity feature sub-information on multiple different intensity evaluation dimensions, so that the intensity feature information can comprehensively characterize the interaction intensity between the object and the video, thereby reducing errors in video recommendation for the object and improving the accuracy of video recommendation for the object.

[0102] In one embodiment, the multiple intensity mapping methods may include a time intensity mapping method, a duration intensity mapping method, and a playback intensity mapping method. Specifically, the step of "performing intensity mapping of attribute information and object interaction feature information corresponding to each video in multiple different dimensions according to the multiple intensity mapping methods to obtain intensity feature sub-information of each video in multiple different dimensions" may include:

[0103] According to the time-intensity mapping method, the attribute information and object interaction feature information corresponding to each video are intensity-mapped to obtain the intensity feature sub-information of each video in the time dimension;

[0104] According to the duration intensity mapping method, the attribute information and object interaction feature information corresponding to each video are intensity mapped to obtain the intensity feature sub-information of each video in the duration dimension;

[0105] According to the playback intensity mapping method, the attribute information and object interaction feature information corresponding to each video are intensity mapped to obtain the intensity feature sub-information of each video in the playback dimension.

[0106] Among them, the time intensity mapping method can be used to convert the attribute information of the video and the object interaction feature information into intensity feature sub-information in the time dimension. For example, the time intensity mapping method can refer to judging the interaction intensity between the object and the video by measuring the distance between the video playback time and the current time. The intensity feature sub-information in the time dimension can refer to information that measures the interaction intensity between the object and the video by the distance between the video playback time and the current time. For example, when the video playback time is closer to the current time, it can be said that the interaction intensity between the object and the video is greater, and the video can more accurately reflect the object's current interest orientation in the video. When the video playback time is farther from the current time, it can be said that the interaction intensity between the object and the video is smaller. By measuring the distance between the video playback time and the current time to judge the interaction intensity of the video, the weight of the intensity feature sub-information of the video with high interaction intensity in the video recommendation can be highlighted, thereby improving the accuracy of the video recommendation.

[0107] Among them, the duration intensity mapping method can refer to judging the interaction intensity between the object and the video by the length of time the user plays the video. For example, the interaction intensity between the object and the video can be judged by the total length of time the user plays a certain video. Among them, the total length can refer to the length of time the user plays the video once, or the total length of time the user plays the video multiple times. Among them, when the total length of time the user plays the video is long, it can be said that the interaction intensity between the object and the video is strong. When the total length of time the user plays the video is short, it can be said that the interaction intensity between the object and the video is weak. The intensity feature sub-information on the duration dimension can refer to information that measures the interaction intensity between the object and the video by the length of time the user plays the video.

[0108] The playback intensity mapping method can refer to determining the interaction strength between an object and a video based on the degree to which a user completes a video. For example, if a user watches a video in its entirety, it can indicate that the user is highly interested in the video and the interaction strength between the object and the video is strong. For another example, if a user only watches the beginning of a video, it can indicate that the user is less interested in the video and the interaction strength between the object and the video is weak. The intensity feature sub-information on the playback dimension can refer to information that measures the interaction strength between an object and a video based on the degree to which a user completes a video.

[0109] In one embodiment, the step of “performing intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the time-intensity mapping method to obtain intensity feature sub-information of each video in the time dimension” may include:

[0110] According to the time-intensity mapping method, the attribute information corresponding to the video is analyzed to obtain the video quantity threshold information, and the object interaction feature information is analyzed to obtain the position information of the video in the video interaction sequence;

[0111] According to the time intensity mapping method, based on the video quantity threshold information, the position information of the video in the video interaction sequence is converted into video reference degree information;

[0112] According to the time intensity mapping method, the video reference degree information is scaled to obtain the intensity feature sub-information of the video in the time dimension.

[0113] Among them, the position information of the video in the video interaction sequence can be used to indicate the position of the video in the video interaction sequence. For example, the video can be added to the video sequence in a preset order, and the time interval between the video's playback time and the current time can be known through the position information of the video in the video interaction sequence. For example, when the video is at the first position in the video interaction sequence, it can be said that the time interval of the video is short, and when the video is at the last position in the video interaction sequence, it can be said that the time interval of the video is long. According to the time interval of the video, the reference value of the video can be reflected. For example, when the time interval of the video is short, it means that the video may be a video that the user has watched recently. At this time, the video has a higher reference value in reflecting the user's video viewing orientation. Conversely, when the time interval of the video is long, it means that the video is a video that the user has watched earlier. At this time, the video has a lower reference value in reflecting the user's video viewing orientation.

[0114] The video quantity threshold information may refer to the total number of videos stored in the video interaction sequence.

[0115] The video reference level information can be used to provide feedback on the reference value of the video.

[0116] In one embodiment, attribute information corresponding to a video can be parsed according to a time-intensity mapping method to obtain video quantity threshold information. In addition, object interaction feature information can also be parsed according to a time-intensity mapping method to obtain video position information in a video interaction sequence.

[0117] For example, the attribute information corresponding to a video may include multiple pieces of information, such as the video's category information, tag information, video duration information, and so on. The video time-intensity mapping method can be used to indicate which information in the attribute information to extract from the video recommendation device. Therefore, the video recommendation transposition can traverse the information included in the video attribute information based on the time-intensity mapping method to obtain the video quantity threshold information.

[0118] For another example, object interaction feature information can include multiple pieces of information. For example, the object interaction feature information can include the actual playback time of the video the subject watched, the video's position in the video interaction sequence, and so on. The video time-intensity mapping method can be used to instruct the video recommendation device which information in the object interaction feature information to extract. Therefore, the video recommendation transposition can traverse the information included in the object interaction feature information based on the time-intensity mapping method to obtain the video's position in the video interaction sequence.

[0119] In one embodiment, the position information of the video in the video interaction sequence may be converted into the video reference degree information based on the video quantity threshold information according to the time intensity mapping method.

[0120] For example, the video quantity threshold information and the position information of the video in the video interaction sequence can be calculated to obtain the video reference level information. For example, the video quantity threshold information minus the position information of the video in the video interaction sequence can be used to obtain the video reference level information, and so on. For example, the video reference level information can be expressed as follows:

[0121] c=L-i+1

[0122] Wherein, c may represent the video reference degree information, L may represent the video quantity threshold information, and i may represent the position information of the video in the video interaction sequence.

[0123] In one embodiment, the number of videos in a video interaction sequence may be large, which may result in a large difference in the video reference level information of videos with different position information. However, in some cases, the reference value of the videos in the video interaction sequence may be relatively high, and the difference in reference value between videos is not large. For example, in the application scenario of short videos, a user may be able to watch several short videos in a short period of time. For example, a user may watch 100 short videos in 1 hour. The reference values ​​of these 100 short videos are theoretically similar, but due to the position information of the short videos in the video interaction sequence, the reference level information of these 100 short videos may vary greatly. Therefore, the video reference level information can be scaled according to the time intensity mapping method, thereby weakening the difference in the video reference level information of the videos, so that while the difference in the reference value of the videos is reflected, the difference will not be too large, thereby ensuring the reliability of video recommendations.

[0124] There are various ways to process the video reference level information to obtain the video's intensity feature information in the temporal dimension. For example, the video reference level information can be normalized to obtain the video's intensity feature sub-information in the temporal dimension. Another example is performing a logarithmic operation on the video reference level information to obtain the video's intensity feature sub-information in the temporal dimension, and so on.

[0125] In one embodiment, the time intensity mapping method may be as follows:

[0126] k i =log(L-i+1)

[0127] Among them, k i It can represent the intensity feature sub-information of the video in the time dimension, L can represent the video quantity threshold information, and i can represent the position information of the video in the video interaction sequence.

[0128] In one embodiment, the step of "performing intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the duration intensity mapping method to obtain intensity feature sub-information of each video in the duration dimension" may include:

[0129] According to the duration-intensity mapping method, the attribute information corresponding to the video is parsed to obtain the duration threshold information, and the object interaction feature information corresponding to the video is parsed to obtain the actual playback time information;

[0130] According to the duration intensity mapping method, the actual playback duration information is compared with the duration threshold information to obtain the target duration information;

[0131] According to the duration intensity mapping method, the target duration information is normalized based on the duration threshold information to obtain the intensity feature sub-information of the video in the duration dimension.

[0132] In one embodiment, the duration intensity mapping method may refer to judging the interaction intensity between an object and a video by the duration of the user playing the video. Therefore, according to the duration intensity mapping method, the attribute information corresponding to the video may be parsed to obtain the duration threshold information, and the object interaction feature information corresponding to the video may be parsed to obtain the actual playback time information.

[0133] The actual play time information may refer to the actual viewing time of the subject when watching the video. For example, if the video is 12 minutes long and the subject stops watching the video after 2 minutes, the actual play time of the video is 2 minutes.

[0134] Among them, the duration threshold information is a filtering value for the actual playback time information, and the duration threshold information can be used to filter out abnormal actual playback time information. For example, the duration threshold information can be used to filter out some actual playback time information with particularly large values. For example, when the subject falls asleep while watching a video, the video may be played in a loop, resulting in the actual playback time information of the video being very large. However, at this time, the actual playback time information of the video has no reference value, so the actual playback time information of the video can be filtered out through the duration threshold information to avoid abnormal information mapping the video recommendation.

[0135] In one embodiment, the actual playback duration information and the duration threshold information can be compared according to the duration intensity mapping method to obtain the target duration information, thereby filtering out some abnormal actual playback duration information. For example, if the actual playback duration information is less than or equal to the duration threshold information, it can be indicated that there is no abnormality in the actual playback duration information, so the actual playback duration information can be determined as the target duration information. For another example, if the actual playback duration information is greater than the duration threshold information, it can be indicated that there is an abnormality in the actual playback duration information, so the duration threshold information can be determined as the target duration information.

[0136] In one embodiment, to facilitate subsequent calculations and improve the efficiency of video recommendation, the target duration information can be normalized based on the duration threshold information according to the duration intensity mapping method to obtain the intensity feature sub-information of the video in the duration dimension. For example, the target duration information can be divided by the duration threshold information to obtain the intensity feature sub-information of the video in the duration dimension.

[0137] In one embodiment, the duration intensity mapping method may be as follows:

[0138]

[0139] Among them, t i It can refer to the actual playback time information of the i-th video in the video interaction sequence, tmax can refer to the duration threshold information, T i It can refer to the intensity feature sub-information of the i-th video in the duration dimension in the video interaction sequence.

[0140] In one embodiment, the step of "performing intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the playback intensity mapping method to obtain intensity feature sub-information of each video in the playback dimension" may include:

[0141] According to the playback intensity mapping method, the attribute information corresponding to the video is parsed to obtain the playback progress threshold information and video duration information, and the object interaction feature information corresponding to the video is parsed to obtain the actual playback time information;

[0142] According to the playback intensity mapping method, the video duration information and the actual playback information of the video are subjected to an arithmetic operation to obtain the playback information after the operation;

[0143] According to the playback intensity mapping method, the calculated playback information and the playback progress threshold information are compared to obtain the intensity feature sub-information of the video in the playback dimension.

[0144] In one embodiment, the playback intensity mapping method can be used to determine the interaction intensity between an object and a video based on the degree to which a user has completed the video. Therefore, the attribute information corresponding to the video can be parsed to obtain the playback progress threshold information and video duration information for the video. Furthermore, the object interaction feature information corresponding to the video can also be parsed to obtain the actual playback time information.

[0145] The video duration information may be used to refer to the time required to play the video. For example, if the video is played for 1 hour, the video duration information of the video is 1 hour.

[0146] The playback progress threshold information can be used to filter outliers, thereby preventing outliers from affecting the results of video recommendations.

[0147] The calculated playback information can be used to illustrate the video's playback completion rate. For example, a high video playback completion rate can indicate that the video has a high reference value, while a low video playback completion rate can indicate that the video has a low reference value.

[0148] In one embodiment, in order to obtain the degree of completion of a video played by a user, an arithmetic operation can be performed on the video duration information and the actual playback information of the video according to a playback intensity mapping method to obtain playback information after the operation. By performing an arithmetic operation on the video duration information and the actual playback information of the video, the difference between the video duration information and the actual playback information can be determined, and thus the user's interest in the video can be determined through the playback information after the operation. For example, the video duration information and the actual playback information can be divided to obtain the playback information after the operation. For another example, the video duration information and the actual playback information can be subtracted and the absolute value can be calculated to obtain the playback information after the operation.

[0149] In one embodiment, in order to avoid the occurrence of numerical anomalies, the post-calculation playback information and the playback progress threshold information can be compared according to the playback intensity mapping method to obtain the intensity characteristic sub-information of the video in the playback dimension. For example, when the post-calculation playback information is greater than the playback progress threshold information, it can be indicated that there is an anomaly in the post-calculation playback information, and the playback progress threshold information can be determined as the intensity characteristic sub-information of the video in the playback dimension. For another example, when the post-calculation playback information is less than or equal to the playback progress threshold information, it can be indicated that there is no anomaly in the post-calculation playback information, and the post-calculation playback information can be determined as the intensity characteristic sub-information of the video in the playback dimension.

[0150] In one embodiment, the playback intensity mapping method may be as follows:

[0151]

[0152] Among them, fr i It can refer to the intensity feature sub-information of the i-th video playback dimension. i It can refer to the actual playback time information of the i-th video in the video interaction sequence. i ) may refer to the video duration information of the i-th video in the video interaction sequence. frmax may refer to the playback progress threshold information.

[0153] In one embodiment, after obtaining the intensity feature sub-information of each video in multiple different dimensions, the intensity feature sub-information of each video in multiple different dimensions may be fused to obtain intensity feature information corresponding to each video. Specifically, the step of "fusion processing the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature information corresponding to each video" may include:

[0154] Encoding the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature encoding information of each video in multiple different dimensions;

[0155] Determine the fusion parameters corresponding to the intensity feature encoding information of each dimension of the video;

[0156] Based on the fusion parameters corresponding to the intensity feature coding information of each dimension, the intensity feature coding information of the video in multiple different dimensions is fused to obtain the intensity feature information corresponding to each video.

[0157] In one embodiment, when recommending videos to an object based on intensity feature information, both the intensity feature information and the video's attribute information need to be encoded into the same mathematical space. This allows the determination of whether a video is of interest to the object to be made within the same mathematical space. Therefore, the intensity feature sub-information of each video in multiple different dimensions can be encoded to obtain encoded intensity feature information for each video in multiple different dimensions.

[0158] For example, artificial intelligence technology can be used to encode the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature encoding information of each video in multiple different dimensions.

[0159] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also involves studying the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0160] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0161] Machine Learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of artificial intelligence. Machine learning and deep learning typically include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and learning by demonstration. Reinforcement learning is a field within machine learning that emphasizes how to act based on the environment to maximize expected benefits. Deep reinforcement learning combines deep learning and reinforcement learning, applying deep learning techniques to solve reinforcement learning problems.

[0162] For example, the intensity feature sub-information of each video in multiple different dimensions can be encoded using convolutional neural networks (CNN), deconvolutional neural networks (DN), deep neural networks (DNN), deep convolutional inverse graphics networks (DCIGN), region-based convolutional networks (RCNN), self-attentive sequential recommendation model (SASRec), fast region-based convolutional networks (Faster RCNN) or bidirectional encoder representations from transformers (BERT) model to obtain the intensity feature encoding information of each video in multiple different dimensions.

[0163] For example, a dual-tower DNN network can be used to encode the intensity feature sub-information of each video in multiple different dimensions to obtain the intensity feature encoding information of each video in multiple different dimensions.

[0164] In one embodiment, for different objects, the intensity feature encoding information of some dimensions may be more important, while that of other dimensions may be less important. Therefore, the proportion of the intensity feature encoding information in the intensity feature information can be adjusted using fusion parameters. Therefore, the fusion parameters corresponding to the intensity feature encoding information of each dimension of the video can be determined.

[0165] In one embodiment, after determining the fusion parameters corresponding to the intensity feature encoding information of each dimension of the video, the intensity feature sub-information of the video in multiple different dimensions can be fused based on the fusion parameters corresponding to the intensity feature encoding information of each dimension to obtain the intensity feature information corresponding to each video.

[0166] For example, the intensity feature encoding information and its corresponding fusion parameter may be multiplied to obtain the multiplied intensity feature encoding information. Then, the multiplied intensity feature encoding information of each dimension may be added to obtain the intensity feature information.

[0167] For example, the intensity feature information of the i-th video in a video interaction sequence can be expressed as follows:

[0168] s i =alpha1*T i +alpha2*fr i +alpha3*k i

[0169] Among them, s i It can represent the intensity feature information of the i-th video, and alpha1, alpha2 and alpha3 can refer to fusion parameters.

[0170] For example, Figure 3 As shown in FIG, it can be a schematic diagram of the generation of the video intensity feature sequence of the object. Figure 3 It can be seen that the intensity feature information of the object may include intensity feature information corresponding to each video.

[0171] 103. Perform multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object.

[0172] In one embodiment, if videos are recommended to an object based solely on intensity feature information, the video recommendation device may only recommend videos of the same category to the object. However, objects often have diverse interests, and recommending only videos of the same category can easily lead to aesthetic fatigue. Therefore, the attribute information corresponding to each video in the object-video interaction sequence can be used to perform multi-interest mapping to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object.

[0173] The interest feature information may be used to illustrate the object's interest orientation in the video. For example, the interest feature information may be the distribution of the object's interest orientation in the video.

[0174] In one embodiment, the attribute information of a video may include category information and tag information of the video. Therefore, interest feature information for an object may be generated based on the category information and tag information of the video. Specifically, the step of "performing multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, wherein the multi-interest feature sequence includes at least one interest feature information corresponding to the object" may include:

[0175] Parse the attribute information corresponding to each video in the video interaction sequence to obtain the category information and label information corresponding to each video;

[0176] Multi-interest mapping is performed on the category information and tag information corresponding to each video to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object.

[0177] The category information of the video can be used to describe the type of the video. For example, the category information of the video can be cute pets, funny, life sharing, technology, or sports, etc.

[0178] The video tag information can abstractly define the content of the video. For example, if the content of the video is related to a TV series, the video tag information can be the title of the TV series. For another example, if the content of the video is related to a person, the video tag information can be the name of the person, and so on.

[0179] In one embodiment, a video's category information may include at least two associated category sub-information items. The association between the category sub-information items may be hierarchical. For example, a video's category may be divided into multiple levels, where the first level is a broad classification, the second level may be a refinement of the first level's categories, and the third level may be a refinement of the second level's categories. For example, a video's category information may be divided into two levels. The first level categorizes videos based on their content, such as comedy, action, romance, fantasy, and so on. The second level may categorize videos based on their region of production. For example, a comedy video may be further refined into Chinese comedy, American comedy, Japanese comedy, and so on. Furthermore, some videos may have diverse genres, and the associated category sub-information may include all the genres covered by the video. For example, if a video has both comedy and romance, its category sub-information may include romance and comedy. Another example is if a video has both fantasy and technology, its category sub-information may include fantasy and technology.

[0180] In one embodiment, when the category information includes at least two associated category sub-information, the step of "performing multi-interest mapping on the category information and tag information corresponding to each video to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object" may include:

[0181] Encode at least two category sub-information and label information of each video to obtain category encoding feature information corresponding to the at least two category sub-information and label encoding feature information corresponding to the label information;

[0182] Determine a fusion parameter corresponding to at least two category encoding feature information and a fusion parameter corresponding to the label encoding feature information;

[0183] Based on the fusion parameters corresponding to at least two category coding feature information of each video and the fusion parameters corresponding to the label coding feature information, the at least two category coding feature information and the label coding feature information corresponding to each video are fused to obtain at least one interest feature information corresponding to the object.

[0184] In one embodiment, artificial intelligence technology can be used to encode at least two category sub-information and label information of each video to obtain category encoding feature information corresponding to the at least two category sub-information and label encoding feature information corresponding to the label information. For example, a dual-tower DNN network can be used to encode at least two category sub-information and label information of each video to obtain category encoding feature information corresponding to the at least two category sub-information and label encoding feature information corresponding to the label information.

[0185] In one embodiment, the focus of information is different for different objects. For example, for some objects, category coding feature information is more important to these objects. And for some objects, label coding feature information is more important to these objects. For example, when some objects watch a video, they will pay attention to the label information carried by the video. At this time, for these objects, label coding feature information can better reflect the object's interest. Therefore, fusion parameters corresponding to at least two category coding feature information and fusion parameters corresponding to label coding feature information can be determined, and then fusion processing can be performed based on the fusion parameters.

[0186] Specifically, the step of “determining fusion parameters corresponding to at least two category encoding feature information and fusion parameters corresponding to label encoding feature information” may include:

[0187] Get object attribute information of an object;

[0188] Perform feature extraction on object attribute information to obtain object feature information;

[0189] Calculating category similarity information between the object feature information and at least two category encoding feature information, and calculating label similarity information between the object feature information and the label encoding feature information;

[0190] At least two fusion parameters corresponding to the category encoding feature information and a fusion parameter corresponding to the label encoding feature information are generated based on the category similarity information and the label similarity information.

[0191] The object attribute information may be information used to describe the object, for example, the object attribute information may be used to describe the location, gender, age group, etc. of the object.

[0192] In one embodiment, feature extraction can be performed on the object attribute information to obtain object feature information. For example, CNN can be used to extract features from the object attribute information to obtain object feature information. Then, category similarity information between the object feature information and at least two category encoding feature information can be calculated, as well as label similarity information between the object feature information and label encoding features can be calculated. For example, category similarity information and label similarity information can be calculated based on methods such as Euclidean distance or Hamming distance. Then, fusion parameters corresponding to at least two category encoding feature information and fusion parameters corresponding to the label encoding features can be generated based on the category similarity information and label similarity information.

[0193] For example, the category similarity information corresponding to the category encoding feature information and the label similarity information corresponding to the label encoding feature can be sorted from large to small, and the fusion parameter can be generated based on the sorting result. For example, similarity information with a larger value will have a larger corresponding fusion parameter, while similarity information with a smaller value will have a smaller corresponding fusion parameter.

[0194] For example, there are two category encoding feature information and one label encoding feature. The similarity information among the three is that the category similarity information of the first category encoding feature information is greater than the label similarity information of the label encoding feature information, and the label similarity information of the label encoding feature information is greater than the category similarity information of the second category encoding feature information. Therefore, the fusion parameter of the first category encoding feature information can be the largest, the fusion parameter of the label encoding feature information can be the second largest, and the fusion parameter of the second category encoding feature information can be the smallest.

[0195] In one embodiment, after determining the fusion parameters, at least two category coding feature information and label coding feature information corresponding to each video can be fused based on the fusion parameters corresponding to at least two category coding feature information of each video and the fusion parameters corresponding to the label coding feature information to obtain at least one interest feature information corresponding to the object.

[0196] For example, the category encoding feature information can be multiplied by its corresponding fusion parameter to obtain the multiplied category encoding feature information. Similarly, the label encoding feature information can be multiplied by its corresponding fusion parameter to obtain the multiplied label encoding feature information. Then, the multiplied category encoding feature information and the multiplied label encoding feature information are added to obtain the interest feature information.

[0197] For example, suppose there are two category sub-information, where the two category sub-information are in a hierarchical relationship, that is, one of the category sub-information can be a first-level category sub-information, and the other category sub-information can be a second-level category sub-information. Then the i-th interest feature information in the multi-interest feature sequence can be expressed as follows:

[0198] mi =beta1*c i +beta2*C i +beta3*Tag i

[0199] Among them, m i It can represent the i-th interest feature information. i It can represent the secondary category coding feature information, and beta1 can be the fusion parameter corresponding to the secondary category sub-information. i It can represent the first-level category coding feature information, and beta2 can be the fusion parameter corresponding to the first-level category sub-information. i It can represent the label encoding feature information, and beta3 can represent the fusion parameter corresponding to the label information.

[0200] For example, Figure 4 As shown in FIG, it can be a schematic diagram of generating a sequence of multiple interest features of an object. Figure 4 It can be seen that the multiple interest feature sequences of an object may include at least one interest feature information, and each interest feature information is generated based on attribute information of the video.

[0201] 104. Fuse the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multi-interest feature sequence to obtain video interaction feature information for the object.

[0202] After obtaining the video intensity feature sequence and the multi-interest feature sequence, video interaction feature information can be generated based on the information in the sequences. To preserve the impact of the original information on the video recommendation process and prevent deviations between the interest feature information and intensity feature information from affecting the accuracy of video recommendations, the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multi-interest feature sequence can be fused to obtain object-specific video interaction feature information.

[0203] Specifically, the object interaction feature information in the video interaction sequence can be first encoded to obtain object interaction coding feature information, so that the object interaction coding feature information, intensity feature information and interest feature information can all be mapped to the same mathematical space for fusion.

[0204] Then, the object interaction coding feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multi-interest feature sequence can be correspondingly fused to obtain the video interaction feature information.

[0205] For example, the intensity feature information, the interest feature information, and the object interaction coding feature information may be added together to obtain the video interaction feature information.

[0206] For example, since object interaction coding feature information is raw information, while intensity feature information and interest feature information are obtained through information mining, they can be distinguished during fusion to highlight their differences. Therefore, the intensity feature information, interest feature information, and object interaction coding feature information can be fused according to the following formula:

[0207] X i =theta1*(s i *m i )+theta2*y i

[0208] Among them, X i Theta1 can represent the fusion parameter corresponding to the intensity feature information and the interest feature information. i can represent the i-th object interaction encoding feature information, and theta2 can represent the fusion parameter corresponding to the object interaction encoding feature information. This approach can balance the influence of original information and information obtained through information mining in the video interaction feature information, so that the video interaction feature information can fully reflect the object's preference for the video.

[0209] For example, Figure 5 As shown in FIG, it can be a schematic diagram of the generation of the video interaction feature sequence of the object. Figure 5 It can be seen that the video interaction feature sequence of an object may include at least one piece of video interaction feature information, and each piece of video interaction feature information is generated by fusing object interaction feature information, intensity feature information, and interest feature information.

[0210] 105. Recommend videos to the target audience based on the video interaction feature information.

[0211] In one embodiment, video recommendations may be made based on video interaction feature information. Specifically, the step of "recommending videos based on video interaction feature information" may include:

[0212] Get object attribute information of an object;

[0213] splicing the video interaction feature information and the object attribute information of the object to obtain splicing feature information;

[0214] Perform multi-layer perception processing on the spliced ​​feature information to obtain object feature representation information;

[0215] Recommend videos for objects based on object feature representation information.

[0216] The object attribute information may be information used to describe the object, for example, the object attribute information may be used to describe the location, gender, age group, etc. of the object.

[0217] In one embodiment, the video interaction feature information and the object attribute information of the object may be spliced ​​together to obtain spliced ​​feature information. The spliced ​​feature information is then subjected to multi-layer perceptron processing to obtain object feature representation information. The multi-layer perceptron processing of the spliced ​​feature information may include processing the spliced ​​feature information using a multi-layer perceptron (MLP) to obtain object feature representation information.

[0218] Then, video recommendations can be made for the object based on the object feature representation information. For example, a video recommendation model can be used to recommend videos for the object based on the object feature representation information. For example, the video recommendation model can be a dual-tower DNN network. The dual-tower DNN network extracts features from the attribute information of the candidate videos to obtain video feature information. The network then calculates the distance between the video feature information and the object feature representation information, and based on this distance, selects videos from the candidate videos to recommend to the object.

[0219] In an embodiment of the present application, a video interaction sequence of an object can be obtained, wherein the video interaction sequence includes attribute information and object interaction feature information corresponding to at least one video; intensity mapping is performed on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video; multi-interest mapping is performed on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, wherein the multi-interest feature sequence includes at least one interest feature information corresponding to the object; the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multi-interest feature sequence are fused to obtain video interaction feature information for the object; and video recommendations are performed for the object based on the video interaction feature information. Since the video interaction feature information combines the intensity feature information, the interest feature information, and the original object interaction feature information, the video interaction feature information can comprehensively feedback the object's interest orientation in the video, thereby improving the accuracy of video recommendations.

[0220] The method described in the above embodiment will be further described in detail below with examples.

[0221] The embodiment of the present application will take the video recommendation method integrated on the server as an example to introduce the method of the embodiment of the present application.

[0222] In one embodiment, if Figure 6 As shown in FIG, a video recommendation method, the specific process is as follows:

[0223] 201. The server obtains a video interaction sequence of an object, where the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information.

[0224] For example, the video interaction sequence of the object is F = [f1, f2, ..., f i ,…,f n ]. Among them, f i It can represent the attribute information and object interaction feature information corresponding to at least one video in a video sequence.

[0225] For example, the attribute information of a video may include the category information, tag information, video duration information, etc. of the video.

[0226] For example, the object interaction feature information of a video may include the actual playback time information of the object watching the video, the position information of the video in the video interaction sequence, and so on.

[0227] 202. The server performs intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence. The video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video.

[0228] In one embodiment, the video attribute information and object interaction feature information can be intensity-mapped along three different dimensions to obtain intensity feature sub-information for the video along the three different dimensions. For example, the three different dimensions may include duration, time, and playback. The intensity feature sub-information along the three different dimensions can then be fused to obtain intensity feature information corresponding to the video.

[0229] In one embodiment, the intensity feature sub-information in the duration dimension may be expressed as follows:

[0230]

[0231] Here, t can represent actual playback time information. For example, t can refer to the playback duration of a video when the object plays it once. Another example is t can refer to the total playback duration of a video when the object plays it multiple times. tmax can represent the duration threshold information. tmax can be defined based on the business to avoid outliers. For example, tmax can be 1 hour, etc.

[0232] In one embodiment, the playback intensity mapping method may be as follows:

[0233]

[0234] Among them, fr i It can refer to the intensity feature sub-information of the i-th video playback dimension. i It can refer to the actual playback time information of the i-th video in the video interaction sequence. i ) may refer to the video duration information of the i-th video in the video interaction sequence. frmax may refer to the playback progress threshold information. frmax may be defined based on the business to avoid outliers.

[0235] In one embodiment, the time intensity mapping method may be as follows:

[0236] k i =log(L-i+1)

[0237] Among them, k i It can represent the intensity feature sub-information of the video in the time dimension, L can represent the video quantity threshold information, and i can represent the position information of the video in the video interaction sequence.

[0238] In one embodiment, the intensity feature information of the i-th video in the video interaction sequence can be expressed as follows:

[0239] s i =alpha1*T i +alpha2*fr i +alpha3*k i

[0240] Among them, s i It can represent the intensity feature information of the i-th video, and alpha1, alpha2 and alpha3 can refer to fusion parameters.

[0241] 203. The server performs multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object.

[0242] For example, suppose there are two category sub-information, where the two category sub-information are in a hierarchical relationship, that is, one of the category sub-information can be a first-level category sub-information, and the other category sub-information can be a second-level category sub-information. Then the i-th interest feature information in the multi-interest feature sequence can be expressed as follows:

[0243] m i =beta1*c i +beta2*C i +beta3*Tag i

[0244] Among them, m i It can represent the i-th interest feature information. i It can represent the secondary category coding feature information, and beta1 can be the fusion parameter corresponding to the secondary category sub-information. i It can represent the first-level category coding feature information, and beta2 can be the fusion parameter corresponding to the first-level category sub-information. i It can represent the label encoding feature information, and beta3 can represent the fusion parameter corresponding to the label information.

[0245] 204. The server fuses the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multi-interest feature sequence to obtain video interaction feature information for the object.

[0246] For example, the intensity feature information, interest feature information, and object interaction coding feature information can be fused according to the following formula:

[0247] X i =theta1*(s i *m i )+theta2*y i

[0248] Among them, X i Theta1 can represent the fusion parameter corresponding to the intensity feature information and the interest feature information. i It can represent the interaction coding feature information of the i-th object, and theta2 can represent the fusion parameter corresponding to the interaction coding feature information of the object.

[0249] 205. The server recommends videos to the target based on the video interaction feature information.

[0250] In an embodiment of the present application, the server obtains a video interaction sequence of an object, wherein the video interaction sequence includes attribute information and object interaction feature information corresponding to at least one video; the server performs intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video; the server performs multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, wherein the multi-interest feature sequence includes at least one interest feature information corresponding to the object; the server fuses the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multi-interest feature sequence to obtain video interaction feature information for the object; the server recommends videos for the object based on the video interaction feature information, thereby improving the accuracy of video recommendations.

[0251] In order to better implement the video recommendation method provided in the embodiment of the present application, a video recommendation device is also provided in one embodiment. The video recommendation device can be integrated into a computer device. The meanings of the terms herein are the same as those in the above-mentioned video recommendation method. For specific implementation details, please refer to the description in the method embodiment.

[0252] In one embodiment, a video recommendation device is provided. The video recommendation device can be integrated into a computer device, such as Figure 7 As shown, the video recommendation device includes: an acquisition unit 301, an intensity mapping unit 302, a multi-interest mapping unit 303, a fusion unit 304 and a video recommendation unit 305, which are specifically as follows:

[0253] An acquiring unit 301 is configured to acquire a video interaction sequence of an object, wherein the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information;

[0254] an intensity mapping unit 302 configured to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video;

[0255] A multi-interest mapping unit 303 is configured to perform multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object;

[0256] a fusion unit 304 configured to fuse the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multiple interest feature sequence to obtain video interaction feature information for the object;

[0257] The video recommendation unit 305 is configured to recommend a video for the object based on the video interaction feature information.

[0258] In one embodiment, the intensity mapping unit 302 may include:

[0259] Mode acquisition subunit, used to obtain multiple intensity mapping modes;

[0260] an intensity mapping subunit, configured to perform intensity mapping of the attribute information and object interaction feature information corresponding to each video in multiple different dimensions according to the multiple intensity mapping modes, to obtain intensity feature sub-information of each video in multiple different dimensions;

[0261] The fusion subunit is used to fuse the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature information corresponding to each video.

[0262] In one embodiment, the intensity mapping subunit may include:

[0263] A first intensity mapping module is configured to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the time intensity mapping method to obtain intensity feature sub-information of each video in the time dimension;

[0264] A second intensity mapping module is configured to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the duration intensity mapping method to obtain intensity feature sub-information of each video in the duration dimension;

[0265] The third intensity mapping module is used to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video according to the playback intensity mapping method to obtain intensity feature sub-information of each video in the playback dimension.

[0266] In one embodiment, the first intensity mapping module may include:

[0267] A first parsing submodule is configured to parse attribute information corresponding to the video according to the time-intensity mapping method to obtain video quantity threshold information, and parse the object interaction feature information to obtain position information of the video in the video interaction sequence;

[0268] a conversion submodule, configured to convert the position information of the video in the video interaction sequence into video reference degree information based on the video quantity threshold information according to the time intensity mapping mode;

[0269] The scaling submodule is configured to scale the video reference degree information according to the time intensity mapping method to obtain intensity feature sub-information of the video in the time dimension.

[0270] In one embodiment, the second intensity mapping module may include:

[0271] A second parsing submodule is configured to parse the attribute information corresponding to the video according to the duration-intensity mapping method to obtain duration threshold information, and parse the object interaction feature information corresponding to the video to obtain actual playback time information;

[0272] A first comparison submodule is configured to compare the actual playback duration information with the duration threshold information according to the duration intensity mapping method to obtain target duration information;

[0273] The normalization submodule is used to normalize the target duration information based on the duration threshold information according to the duration intensity mapping method to obtain the intensity feature sub-information of the video in the duration dimension.

[0274] In one embodiment, the third intensity mapping module may include:

[0275] A third parsing submodule is configured to parse the attribute information corresponding to the video according to the playback intensity mapping method to obtain playback progress threshold information and video duration information, and to parse the object interaction feature information corresponding to the video to obtain actual playback time information;

[0276] an arithmetic operation submodule, configured to perform an arithmetic operation on the video duration information of the video and the actual playback information according to the playback intensity mapping mode to obtain playback information after the operation;

[0277] The second comparison submodule is configured to compare the calculated playback information with the playback progress threshold information according to the playback intensity mapping method to obtain intensity feature sub-information of the video in the playback dimension.

[0278] In one embodiment, the first fusion subunit may include:

[0279] A first encoding module is used to encode the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature encoding information of each video in multiple different dimensions;

[0280] A first parameter determination module is used to determine the fusion parameter corresponding to the intensity feature coding information of each dimension of the video;

[0281] The first fusion module is used to fuse the intensity feature coding information of the video in multiple different dimensions based on the fusion parameters corresponding to the intensity feature coding information of each dimension to obtain intensity feature information corresponding to each video.

[0282] In one embodiment, the multi-interest mapping unit 303 may include:

[0283] A parsing subunit, configured to parse the attribute information corresponding to each video in the video interaction sequence to obtain category information and tag information corresponding to each video;

[0284] The multi-interest mapping subunit is used to perform multi-interest mapping on the category information and tag information corresponding to each video to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object.

[0285] In one embodiment, the multi-interest mapping subunit may include:

[0286] A second encoding module is configured to encode the at least two category sub-information and the label information of each video to obtain category encoding feature information corresponding to the at least two category sub-information and label encoding feature information corresponding to the label information;

[0287] A second parameter determination module is used to determine the fusion parameters corresponding to the at least two category encoding feature information and the fusion parameters corresponding to the label encoding feature information;

[0288] The second fusion module is used to fuse the at least two category coding feature information and label coding feature information corresponding to each video based on the fusion parameters corresponding to the at least two category coding feature information of each video and the fusion parameters corresponding to the label coding feature information to obtain at least one interest feature information corresponding to the object.

[0289] In one embodiment, the second parameter determination module may include:

[0290] Get submodule, used to get object attribute information of an object;

[0291] A feature extraction submodule is used to extract features from the attribute information of the object to obtain object feature information;

[0292] a similarity information calculation submodule, configured to calculate category similarity information between the object feature information and the at least two category coding features, and to calculate label similarity information between the object feature information and the label coding feature;

[0293] The parameter generation submodule is used to generate fusion parameters corresponding to the at least two category coding features and fusion parameters corresponding to the label coding features based on the category similarity information and the label similarity information.

[0294] In one embodiment, the video recommendation unit 305 may include:

[0295] An information acquisition subunit, used to acquire object attribute information of an object;

[0296] a splicing subunit, configured to splice the video interaction feature information and the object attribute information of the object to obtain splicing feature information;

[0297] a multi-layer perception processing subunit, configured to perform multi-layer perception processing on the splicing feature information to obtain object feature representation information;

[0298] The video recommendation subunit is used to recommend videos for the object based on the object feature representation information.

[0299] In specific implementation, the above units can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units can be found in the previous method embodiments and will not be repeated here.

[0300] The accuracy of video recommendation can be improved by the above-mentioned video recommendation device.

[0301] The embodiment of the present application also provides a computer device, which may include a terminal or a server. For example, the computer device may be used as a video recommendation terminal, which may be a mobile phone, a tablet computer, etc.; for another example, the computer device may be a server, such as a video recommendation server. Figure 8 As shown, it shows a schematic diagram of the structure of the terminal involved in the embodiment of the present application, specifically:

[0302] The computer device may include one or more processing core processors 401, one or more computer readable storage media memories 402, a power supply 403, an input unit 404 and other components. Those skilled in the art will understand that Figure 8 The computer device structure shown in the figure does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently.

[0303] Processor 401 is the control center of the computer device, connecting the various components of the entire computer device using various interfaces and circuits. It executes the various functions of the computer device and processes data by running or executing software programs and / or modules stored in memory 402 and accessing data stored in memory 402. Optionally, processor 401 may include one or more processing cores; preferably, processor 401 may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interfaces, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 401.

[0304] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 402 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0305] The computer device also includes a power supply 403 for supplying power to various components. Preferably, the power supply 403 can be logically connected to the processor 401 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The power supply 403 can also include one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components.

[0306] The computer device may further include an input unit 404, which may be configured to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.

[0307] Although not shown, the computer device may further include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 401 in the computer device will load the executable files corresponding to one or more application processes into the memory 402 according to the following instructions, and the processor 401 will run the application stored in the memory 402 to implement various functions as follows:

[0308] Acquire a video interaction sequence of an object, wherein the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information;

[0309] Performing intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video;

[0310] Performing multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object;

[0311] fusing the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multiple interest feature sequences to obtain video interaction feature information for the object;

[0312] Video recommendations are performed for the object based on the video interaction feature information.

[0313] The specific implementation of the above operations can be found in the previous embodiments and will not be repeated here.

[0314] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0315] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by a computer program, or by controlling related hardware through a computer program. The computer program may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0316] To this end, an embodiment of the present application further provides a storage medium storing a computer program that can be loaded by a processor to execute the steps of any of the video recommendation methods provided in the embodiments of the present application. For example, the computer program can execute the following steps:

[0317] Acquire a video interaction sequence of an object, wherein the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information;

[0318] Performing intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video;

[0319] Performing multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object;

[0320] fusing the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multiple interest feature sequences to obtain video interaction feature information for the object;

[0321] Video recommendations are performed for the object based on the video interaction feature information.

[0322] The specific implementation of the above operations can be found in the previous embodiments and will not be repeated here.

[0323] Since the computer program stored in the storage medium can execute the steps of any video recommendation method provided in the embodiments of the present application, the beneficial effects that can be achieved by any video recommendation method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0324] The above is a detailed introduction to a video recommendation method, device, computer equipment and storage medium provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A video recommendation method, characterized in that: include: Acquire a video interaction sequence of an object, wherein the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information; Performing intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video; Performing multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object; fusing the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multiple interest feature sequences to obtain video interaction feature information for the object; Video recommendations are performed for the object based on the video interaction feature information.

2. The method according to claim 1, characterized in that The intensity mapping is performed on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video, including: Get multiple intensity mapping methods; According to the multiple intensity mapping methods, the attribute information and object interaction feature information corresponding to each video are subjected to intensity mapping in multiple different dimensions to obtain intensity feature sub-information of each video in multiple different dimensions; The intensity feature sub-information of each video in multiple different dimensions is fused to obtain intensity feature information corresponding to each video.

3. The method according to claim 2, characterized in that The multiple intensity mapping methods include a time intensity mapping method, a duration intensity mapping method, and a playback intensity mapping method. According to the multiple intensity mapping methods, attribute information and object interaction feature information corresponding to each video are subjected to intensity mapping in multiple different dimensions to obtain intensity feature sub-information of each video in multiple different dimensions, including: According to the time-intensity mapping method, intensity mapping is performed on the attribute information and object interaction feature information corresponding to each video to obtain intensity feature sub-information of each video in the time dimension; According to the duration intensity mapping method, intensity mapping is performed on the attribute information and object interaction feature information corresponding to each video to obtain intensity feature sub-information of each video in the duration dimension; According to the playback intensity mapping method, the attribute information and object interaction feature information corresponding to each video are intensity mapped to obtain the intensity feature sub-information of each video in the playback dimension.

4. The method according to claim 3, characterized in that According to the time-intensity mapping method, intensity mapping is performed on the attribute information and object interaction feature information corresponding to each video to obtain intensity feature sub-information of each video in the time dimension, including: Parsing attribute information corresponding to the video according to the time-intensity mapping method to obtain video quantity threshold information, and parsing the object interaction feature information to obtain position information of the video in the video interaction sequence; According to the time intensity mapping method, based on the video quantity threshold information, the position information of the video in the video interaction sequence is converted into video reference degree information; According to the time intensity mapping method, the video reference degree information is scaled to obtain the intensity feature sub-information of the video in the time dimension.

5. The method according to claim 3, characterized in that According to the duration intensity mapping method, intensity mapping is performed on the attribute information and object interaction feature information corresponding to each video to obtain intensity feature sub-information of each video in the duration dimension, including: According to the duration intensity mapping method, the attribute information corresponding to the video is parsed to obtain duration threshold information, and the object interaction feature information corresponding to the video is parsed to obtain actual playback duration information; Comparing the actual playback duration information with the duration threshold information according to the duration intensity mapping method to obtain target duration information; According to the duration intensity mapping method, the target duration information is normalized based on the duration threshold information to obtain intensity feature sub-information of the video in the duration dimension.

6. The method according to claim 3, characterized in that According to the playback intensity mapping method, the attribute information and object interaction feature information corresponding to each video are intensity mapped to obtain the intensity feature sub-information of each video in the playback dimension, including: According to the playback intensity mapping method, the attribute information corresponding to the video is parsed to obtain playback progress threshold information and video duration information, and the object interaction feature information corresponding to the video is parsed to obtain actual playback duration information; performing an arithmetic operation on the video duration information of the video and the actual playback duration information according to the playback intensity mapping method to obtain playback information after the operation; According to the playback intensity mapping method, the calculated playback information is compared with the playback progress threshold information to obtain the intensity feature sub-information of the video in the playback dimension.

7. The method according to claim 2, characterized in that The fusing of the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature information corresponding to each video includes: Encoding the intensity feature sub-information of each video in multiple different dimensions to obtain intensity feature encoding information of each video in multiple different dimensions; Determine the fusion parameters corresponding to the intensity feature encoding information of each dimension of the video; Based on the fusion parameters corresponding to the intensity feature coding information of each dimension, the intensity feature coding information of the video in multiple different dimensions is fused to obtain the intensity feature information corresponding to each video.

8. The method according to claim 1, characterized in that The performing multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, wherein the multi-interest feature sequence includes at least one interest feature information corresponding to the object, including: Parsing the attribute information corresponding to each video in the video interaction sequence to obtain category information and tag information corresponding to each video; Multi-interest mapping is performed on the category information and tag information corresponding to each video to obtain a multi-interest feature sequence, where the multi-interest feature sequence includes at least one interest feature information corresponding to the object.

9. The method according to claim 8, characterized in that The category information includes at least two associated category sub-information; the multi-interest mapping is performed on the category information and tag information corresponding to each video to obtain a multi-interest feature sequence, wherein the multi-interest feature sequence includes at least one interest feature information corresponding to the object, including: Encoding the at least two category sub-information and the tag information of each video to obtain category coding feature information corresponding to the at least two category sub-information and tag coding feature information corresponding to the tag information; Determining a fusion parameter corresponding to the at least two category encoding feature information and a fusion parameter corresponding to the label encoding feature information; Based on the fusion parameters corresponding to at least two category coding feature information of each video and the fusion parameters corresponding to the label coding feature information, the at least two category coding feature information and the label coding feature information corresponding to each video are fused to obtain at least one interest feature information corresponding to the object.

10. The method according to claim 9, characterized in that The determining of the fusion parameters corresponding to the at least two category encoding feature information and the fusion parameters corresponding to the label encoding feature information includes: Get object attribute information of an object; performing feature extraction on the object attribute information to obtain object feature information; Calculating category similarity information between the object feature information and the at least two category encoding feature information, and calculating label similarity information between the object feature information and the label encoding feature information; Based on the category similarity information and the label similarity information, fusion parameters corresponding to the at least two category encoding feature information and fusion parameters corresponding to the label encoding feature information are generated.

11. The method according to claim 1, wherein The performing of video recommendation for the object based on the video interaction feature information includes: Get object attribute information of an object; splicing the video interaction feature information and the object attribute information of the object to obtain splicing feature information; Performing multi-layer perception processing on the splicing feature information to obtain object feature representation information; Recommend videos for the object based on the object feature representation information.

12. A video recommendation device, characterized in that: include: an acquiring unit, configured to acquire a video interaction sequence of an object, wherein the video interaction sequence includes attribute information corresponding to at least one video and object interaction feature information; an intensity mapping unit, configured to perform intensity mapping on the attribute information and object interaction feature information corresponding to each video in the video interaction sequence to obtain a video intensity feature sequence, wherein the video intensity feature sequence includes intensity feature information corresponding to each video, and the intensity feature information represents the interaction intensity between the object and each video; a multi-interest mapping unit, configured to perform multi-interest mapping on the attribute information corresponding to each video in the video interaction sequence to obtain a multi-interest feature sequence, wherein the multi-interest feature sequence includes at least one interest feature information corresponding to the object; a fusion unit, configured to fuse the object interaction feature information in the video interaction sequence, the intensity feature information in the video intensity feature sequence, and the interest feature information in the multiple interest feature sequence to obtain video interaction feature information for the object; A video recommendation unit is used to recommend videos for the object based on the video interaction feature information.

13. A computer device, characterized in that: The system comprises a memory and a processor; the memory stores an application program, and the processor is configured to run the application program in the memory to perform the operations in the video recommendation method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the steps in the video recommendation method according to any one of claims 1 to 11.

15. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the video recommendation method according to any one of claims 1 to 11 are implemented.

Images