CN115391663A - Video recommendation method and device, computer equipment and storage medium - Google Patents

Abstract

The present application relates to a video recommendation method, device, computer equipment and storage medium. The method includes: acquiring a video viewing set corresponding to the target user identifier, the video viewing set includes at least two video viewing sequences with different video durations, extracting corresponding feature vectors based on each video viewing sequence, and fusing each feature vector to form Fusion vectors, according to the obtained inner product results of each video vector to be recommended and the fusion vector, determine the recommended value of each video vector to be recommended, and push the corresponding video vector corresponding to the video vector to be recommended according to the descending order of the recommended value The video data is sent to the terminal corresponding to the target user identification. Since the above method not only extracts feature vectors from different video viewing sequences to capture the user's preferences, but also combines the links between different video viewing sequences, so as to fully and accurately capture the user's interest in the video. The interest preference of content optimizes the push results of video content.

Description

Video recommendation method, device, computer equipment and storage medium

technical field

The present application relates to the field of computer technology, and in particular to a video recommendation method, device, computer equipment and storage medium.

Background technique

With the development of video software, users can watch long videos or short videos according to their preferences through video software. The video software will capture the user's preferences for videos based on the user's historical viewing behavior, and thus push the video content that the user may be interested in. Viewing sequences are divided into long video viewing sequences, medium video viewing sequences, and short video viewing sequences. Currently, existing video push methods model and analyze video content that users may be interested in for different types of viewing sequences, ignoring different Therefore, it cannot fully and accurately capture user interest preferences, resulting in suboptimal push results.

Contents of the invention

In order to solve the above technical problems, the present application provides a video recommendation method, device, computer equipment and storage medium.

In the first aspect, the present application provides a video recommendation method, including:

Obtain a video viewing set corresponding to the target user identifier, wherein the video viewing set includes at least two video viewing sequences with different video durations;

Extracting corresponding feature vectors based on each of the video viewing sequences, and fusing each of the feature vectors to form a fusion vector;

According to the obtained inner product result of each video vector to be recommended and the fusion vector, the recommendation value of each video vector to be recommended is determined, wherein the video vector to be recommended is used to indicate the attribute information of the video to be recommended;

Pushing the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommendation value.

In a second aspect, the present application provides a video recommendation device, including:

An acquisition module, configured to acquire a video viewing set corresponding to the target user identifier, wherein the video viewing set includes at least two video viewing sequences with different video durations;

A fusion module, configured to extract corresponding feature vectors based on each of the video viewing sequences, and fuse each of the feature vectors to form a fusion vector;

A determination module, configured to determine the recommended value of each video vector to be recommended according to the acquired inner product result of each video vector to be recommended and the fusion vector, wherein the video vector to be recommended is used to indicate the video to be recommended attribute information;

The push module is configured to push the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommendation value.

In a third aspect, the present application provides a computer device, including a memory, a processor, and a computer program stored on the memory and operable on the processor. When the processor executes the computer program, the following steps are implemented:

Obtain a video viewing set corresponding to the target user identifier, wherein the video viewing set includes at least two video viewing sequences with different video durations;

Extracting corresponding feature vectors based on each of the video viewing sequences, and fusing each of the feature vectors to form a fusion vector;

According to the obtained inner product result of each video vector to be recommended and the fusion vector, the recommendation value of each video vector to be recommended is determined, wherein the video vector to be recommended is used to indicate the attribute information of the video to be recommended;

Pushing the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommendation value.

In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

Obtain a video viewing set corresponding to the target user identifier, wherein the video viewing set includes at least two video viewing sequences with different video durations;

Extracting corresponding feature vectors based on each of the video viewing sequences, and fusing each of the feature vectors to form a fusion vector;

According to the obtained inner product result of each video vector to be recommended and the fusion vector, the recommendation value of each video vector to be recommended is determined, wherein the video vector to be recommended is used to indicate the attribute information of the video to be recommended;

Pushing the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommendation value.

Based on the above video recommendation method, the video viewing set corresponding to the target user identifier is obtained, the video viewing set includes at least two video viewing sequences with different video durations, the corresponding feature vectors are respectively extracted based on each of the video viewing sequences, and each of the video viewing sequences is extracted The above feature vectors are fused to form a fusion vector, that is, user preferences are mined from different video viewing sequences and connections between different video viewing sequences are established. According to the obtained inner product results of each video vector to be recommended and the fusion vector, each The recommendation value of the video vector to be recommended, wherein, the video vector to be recommended is used to indicate the attribute information of the video to be recommended, and is predicted by using a fusion vector fused with corresponding information of different video viewing sequences, and the target user identifies the corresponding user for each The preference degree of the video content corresponding to the video vector to be recommended, that is, the recommendation value of each video vector to be recommended is obtained, and the video data corresponding to the video vector to be recommended is pushed to the terminal corresponding to the target user identification according to the descending order of the recommended value. The above method not only extracts feature vectors from different video viewing sequences to capture user preferences, but also combines the connections between different video viewing sequences, so as to fully and accurately capture user preferences for video content and optimize the push of video content result.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description serve to explain the principles of the invention.

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, In other words, other drawings can also be obtained from these drawings without paying creative labor.

Fig. 1 is an application environment diagram of a video recommendation method in an embodiment;

Fig. 2 is a schematic flow chart of a video recommendation method in an embodiment;

Fig. 3 is a structural block diagram of a video recommendation device in an embodiment;

Figure 4 is an internal block diagram of a computer device in one embodiment.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments It is a part of the embodiments of this application, but not all of them. Based on the embodiments in the present application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present application.

Fig. 1 is an application environment diagram of a video recommendation method in an embodiment. Referring to FIG. 1 , the video recommendation method is applied to a video recommendation system. The video recommendation system includes a terminal 110 and a server 120 . The terminal 110 and the server 120 are connected through the network, and the video playing software is installed in the terminal 110, and the user can log in the video playing software through the user ID to watch the video content. The terminal 110 may specifically be a desktop terminal or a mobile terminal, and the mobile terminal may specifically be at least one of a mobile phone, a tablet computer, a notebook computer, and the like. The server 120 can be implemented by an independent server or a server cluster composed of multiple servers.

In an embodiment, FIG. 2 is a schematic flowchart of a video recommendation method in an embodiment. Referring to FIG. 2 , a video recommendation method is provided. This embodiment is mainly illustrated by taking the method applied to the server 120 in the above-mentioned FIG. 1 as an example. The video recommendation method specifically includes the following steps:

Step S210, acquiring a video viewing set corresponding to the target user identifier, wherein the video viewing set includes at least two video viewing sequences with different video durations.

Specifically, the target user ID is any user ID used to log in to the video playback software. The user ID specifically includes the user's identity information and custom characters. The identity information includes phone numbers, email addresses, or third-party application accounts. The custom characters can be It is composed of one or more combinations of any number, letter, symbol, etc. Different user IDs are used to indicate different users, and different users have different preferences for different video content, so the video viewing sets corresponding to different user IDs are also different, and the video viewing sets include at least two video viewing sequences with different video durations According to the length of the video, the video viewing sequence is divided into long video viewing sequence, medium video viewing sequence, and short video viewing sequence. Usually, a video with a video duration of more than 20 minutes is defined as a long video, and a video with a video duration of less than 1 minute is defined as a short video. For videos, videos with a video duration between 1 and 20 minutes are defined as medium videos, and the definition of video duration for long videos, medium videos, and short videos can also be customized. In this embodiment, the video viewing set includes a long video viewing sequence and a short video viewing sequence.

Video viewing sequences of different video lengths contain video information corresponding to corresponding video durations, that is, long video viewing sequences include multiple video description vectors corresponding to long videos, medium video viewing sequences include multiple video description vectors corresponding to medium videos, and short video viewing sequences include multiple video description vectors corresponding to medium videos. A sequence includes video description vectors corresponding to multiple short videos. The video description vector contains the attribute information of the video, and the attribute information includes video ID (Identity document, logo), video album ID, drama name, type, channel ID, solo label, participating actors, director, number of actor works, number of director works , screenwriter, all award-winning names of the play, protagonist name, movie rating, movie restriction rating and description, global release time, China release time, page first online time, etc., that is, the attribute information includes description information of multiple dimensions of the video.

Step S220, extract corresponding feature vectors based on each of the video viewing sequences, and fuse each of the feature vectors to form a fusion vector.

Specifically, a feature vector is extracted based on each video viewing sequence, and the feature vector is used to indicate the user's preference vector for the video, and then the feature vectors corresponding to each video viewing sequence are fused to form a fusion vector, that is, from different video viewing sequences In this method, user preference information is mined and connections between different video viewing sequences are established, so that the video viewing sequences of different video durations are fully interacted with information.

Step S230: Determine the recommended value of each video vector to be recommended according to the obtained inner product result of each video vector to be recommended and the fusion vector, wherein the video vector to be recommended is used to indicate the attribute of the video to be recommended information.

Specifically, the video vector to be recommended is the description vector of the video to be recommended, and the video to be recommended is a video whose viewing status is unwatched relative to the target user identification in the video database. The video database contains videos with different video durations, that is, the video database includes multiple Long video, medium video, short video. Based on the inner product result of the video vector to be recommended and the fusion vector, the video vector to be recommended and the fusion vector are processed by dot product, and the attribute information of the video to be recommended is effectively combined with the user's interest, so as to obtain the recommendation value of each video vector to be recommended. Arrange in descending order according to the recommendation value of each video vector to be recommended to obtain a roughly sorted video recommendation list.

Step S240, push the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommended value.

Specifically, the higher the recommendation value, it means that the corresponding video to be recommended is more likely to meet the user's video viewing preferences, and the lower the recommendation value, it means that the user's preference for the corresponding video to be recommended is lower. The video data corresponding to the recommended video vector is pushed to the terminal corresponding to the target user identifier, N is a positive integer greater than zero, and N is less than or equal to the number of video vectors to be recommended in the video recommendation list.

Since the above steps not only extract feature vectors from different video viewing sequences to capture the user's preferences, but also combine the connections between different video viewing sequences, it is possible to fully and accurately capture the user's interest preferences for video content, and improve the quality of videos to be recommended. The rough layout effect, that is, the push result of the video content is optimized.

In one embodiment, the acquisition of the video viewing set corresponding to the target user identifier includes:

Obtaining a historical video collection corresponding to the target user identifier, wherein the historical video collection includes at least two historical video sequences with different video durations, and the historical video sequences include a plurality of video attribute vectors;

Performing one-hot encoding processing on each of the historical video sequences to obtain a corresponding video encoding sequence, wherein the video encoding sequence includes a plurality of one-hot encoding vectors corresponding to the video attribute vectors;

Perform dimensionality reduction processing on each of the video coding sequences to obtain the corresponding video viewing sequence, wherein the video viewing sequence includes a plurality of video description vectors.

Specifically, the historical video collection includes historical video sequences of different video durations. The historical video sequences are divided into historical long video sequences, historical medium video sequences, and historical short video sequences according to the video duration. In this embodiment, the historical video collection includes historical long video sequences. Video sequence and historical short video sequence, the historical long video sequence is recorded as L=(l1, l2, ..., ln), wherein, l1～ln are n video attribute vectors in the historical long video sequence, in the historical long video sequence Each video attribute vector of is used to indicate the attribute information of a long video, and the historical short video sequence is recorded as S=(s1, s2, ..., sn), and s1～sn are n video attribute vectors in the historical short video sequence, Each video attribute vector in the historical short video sequence is used to indicate the attribute information of a short video.

The historical long video sequence and the historical short video sequence are respectively subjected to one-hot encoding (one-hot) processing, that is, the encoding of the video attribute vector is converted into a one-hot code vector, so as to obtain the corresponding video encoding sequence, and the historical long video sequence is subjected to one-hot encoding. The long video encoding sequence is obtained, and the short video encoding sequence is obtained through one-hot encoding of the historical short video sequence.

Then each video coding sequence is subjected to dimensionality reduction processing to obtain the video viewing sequence after dimensionality reduction, that is, the video viewing set in step S210 is obtained by the historical video collection through coding dimensionality reduction processing, that is, the long video coding sequence is subjected to dimensionality reduction processing A long video viewing sequence is obtained, and a short video coding sequence is subjected to dimensionality reduction processing to obtain a short video viewing sequence.

In one embodiment, the video viewing sequence includes a first dimensionality reduction sequence and a second dimensionality reduction sequence, and performing dimensionality reduction processing on each of the video encoding sequences to obtain a corresponding video viewing sequence, including at least the following one:

Multiplying each of the video encoding sequences by the first mapping matrix respectively to obtain the corresponding first dimensionality reduction sequence, wherein the first mapping matrix includes matrix parameters corresponding to video attributes;

Multiplying each of the video coding sequences by the second mapping matrix respectively to obtain the corresponding second dimensionality reduction sequence, wherein the second mapping matrix includes matrix parameters corresponding to the preference attributes.

Specifically, the video viewing sequence includes a first dimensionality reduction sequence and a second dimensionality reduction sequence, which represent different dimensionality reduction sequences obtained by using two dimensionality reduction methods for the same video coding sequence. If the first dimensionality reduction method is adopted, the length The video coding sequence and the short video coding sequence are respectively multiplied by the first mapping matrix to realize the weight embedding mapping (embedding), and the corresponding first dimensionality reduction sequence is obtained, that is, the long video coding sequence is multiplied by the first mapping matrix to obtain the long In the first dimensionality reduction sequence of the video, the coding sequence of the short video is multiplied by the first mapping matrix to obtain the first dimensionality reduction sequence of the short video. The first dimensionality reduction sequence of the long video is recorded as C=(c1, c2, ..., cn), where c1～cn are the video description vectors of n long videos in the first dimensionality reduction sequence of the long video, and the short video first The dimensionality reduction sequence is denoted as I=(i1, i2,...,in), wherein, i1～in are the video description vectors of n short videos in the first dimensionality reduction sequence of the short video, and each video description vector not only contains the corresponding video The attribute information of also includes the association relationship between the video and other videos in the same sequence.

If the second dimensionality reduction method is used, the long video coding sequence and the short video coding sequence are multiplied by the second mapping matrix to realize the weight embedding mapping (embedding), and the corresponding second dimensionality reduction sequence is obtained, that is, the long video The coding sequence is multiplied by the second mapping matrix to obtain the second dimensionality reduction matrix of the long video, and the short video coding sequence is multiplied by the second mapping matrix to obtain the second dimensionality reduction matrix of the short video. The first mapping matrix and the second mapping matrix can include the same and/or different matrix parameters. Most of the first mapping matrix are matrix parameters related to video attributes, such as video attribute features such as video type, video ID, and channel ID. Most of the matrix parameters in the two-mapping matrix are matrix parameters related to preference attributes, for example, preference features that users may be interested in, such as play title, participating actors, director, and protagonist name. The second dimensionality reduction sequence of the long video is recorded as D=(d1, d2, ..., dn), wherein, d1～dn is the video description vector of n long videos in the second dimensionality reduction sequence of the long video, and the short video second The dimensionality reduction sequence is denoted as K=(k1, k2,...,kn), where k1˜kn are the video description vectors of n short videos in the second short video dimensionality reduction sequence.

The dimensionality reduction sequence obtained according to the first dimensionality reduction method is used to reflect the user's preference for each long video or short video, and the dimensionality reduction sequence obtained according to the second dimensionality reduction method is used to reflect the user's preference for each preference feature , so as to mine the user's interest characteristics.

In one embodiment, the feature vector includes a characterization vector, the fusion vector includes a characterization fusion vector, and the corresponding feature vectors are respectively extracted based on each of the video viewing sequences, and each of the feature vectors is fused to form a fusion vector, including:

respectively performing mean pooling processing on each of the first dimensionality reduction sequences to obtain the corresponding characterization vectors;

Determine a primary fusion vector according to the dot product results between each of the characterization vectors;

performing fusion processing on the first-level fusion vector and all the first dimensionality reduction sequences to obtain a second-level fusion vector;

Adding the first-level fusion vector and the second-level fusion vector to form the representation fusion vector.

Specifically, the feature vector includes a characterization vector, and the characterization vector is used to indicate the user's preference for different video viewing sequences. Mean pooling (MeanPooling) processing is performed on the first dimensionality reduction sequence of the long video and the first dimensionality reduction sequence of the short video respectively, that is, the average of each video description vector in the first dimensionality reduction sequence of the long video is averaged to obtain the corresponding long video representation vector Ec1, average the video description vectors in the first dimensionality reduction sequence of the short video to obtain the corresponding short video characterization vector Ei1, that is, the characterization vector includes the long video characterization vector and the short video characterization vector, which is used to reflect the user’s preference for long video or short video Likeness of the video.

The dot product result between the long video characterization vector Ec1 and the short video characterization vector Ei1 is used as the first-level fusion vector Ef1, that is, part of the information of the first dimensionality reduction sequence of the long video and the first dimensionality reduction sequence of the short video is fused, and then a The first-level fusion vector Ef1 and the first dimensionality reduction sequence C of the long video and the first dimensionality reduction sequence I of the short video are input into the attention network model (Attention) as input parameters to output the fused level-2 fusion vector Ef2. On the basis of the fusion vector, continue to fuse the first dimensionality reduction sequence C of the long video and the first dimensionality reduction sequence I of the short video to generate a higher-order fusion vector Ef2, and add the first-level fusion vector Ef1 to the second-level fusion vector Ef2 , forming the characterization fusion vector F of the double sequence of the first dimensionality reduction sequence of the long video and the first dimensionality reduction sequence of the short video. Multiple dimensions.

In one embodiment, the feature vector further includes a preference vector, the fusion vector includes a preference fusion vector, and the corresponding feature vectors are respectively extracted based on each of the video viewing sequences, and each of the feature vectors is fused to form Fusion vectors, including:

Based on the association relationship between different video description vectors in each of the second dimensionality reduction sequences, determine a relationship learning vector corresponding to each of the video description vectors, wherein the relationship learning vector includes a target description vector and the target description vector The association relationship with each of the video description vectors in the second dimensionality reduction sequence, the target description vector is any one of the video description vectors in the second dimensionality reduction sequence;

determining a preference vector corresponding to each of the second dimensionality reduction sequences according to the confidence that the plurality of relational learning vectors corresponding to the same second dimensionality reduction sequence belong to each preference feature;

The preference vectors corresponding to each of the second dimensionality reduction sequences are fused to form the preference fusion vector.

Specifically, the feature vector also includes a preference vector, which is used to indicate the user's preference for different preference features, that is, the preference vector reflects the user's degree of interest in different aspects of the video, and the preference features include all or part of the above attribute information Characteristics, such as starring actors, directors, category of video, etc.

The second dimensionality reduction sequence D of the long video and the second dimensionality reduction sequence K of the short video are respectively input into the multi-head self-attention network model as input parameters, so as to describe the relationship between different video description vectors in each second dimensionality reduction sequence After learning, output the relationship learning vector corresponding to each video description vector, that is, the relationship learning vector not only contains the attribute information corresponding to the corresponding video, but also includes the relationship between the video and other videos in the same sequence, that is, the description of each video in the same sequence Vectors are fused and learned, and then multiple relational learning vectors output by the multi-head self-attention network model are input into the fully connected layer network structure as input parameters, and each relational learning vector is classified, and each relational learning vector is calculated to belong to each preference feature According to the confidence degree of each relational learning vector belonging to each preference feature, the user’s preference degree for each preference feature is determined, and the preference degree of each preference feature forms the corresponding preference vector of the second dimensionality reduction sequence, that is, the long video No. The long video preference vector corresponding to the second dimensionality reduction sequence D and the short video preference vector corresponding to the second dimensionality reduction sequence K of the short video.

The long video preference vector is denoted as P=(p1, p2,...,pm), where p1～pm is used to indicate the user's preference for m preference features when facing long videos, and the short video preference vector is denoted as Q =(q1, q2, ..., qm), where q1~qm are used to indicate the user's preference for m preference features when facing end-to-end videos, and different users have the same preference for long or short videos Features correspond to different preferences. For example, the preference feature is actors, and young people prefer movies or TV series played by young actors when facing long videos, that is, they have a higher preference for long videos played by young actors; When faced with long videos, the elderly prefer movies or TV series with senior actors, that is, they have a higher preference for long videos with senior actors. In this way, the preference characteristics of different user identifiers for different videos and the degree of preference for each preference characteristic can be mined.

The long video preference vector and the short video preference vector are input into the MLP network and fused to form a preference fusion vector H. The preference fusion vector is used to reflect the user's preference characteristics for different videos and the degree of preference for each preference characteristic, so as to accurately capture the user's preference for Preference characteristics of the video.

In one embodiment, the determining the recommended value of each video vector to be recommended according to the obtained inner product result of each video vector to be recommended and the fusion vector includes:

converting the acquired attribute information of each video to be recommended into a corresponding vector of the video to be recommended;

According to the inner product result between the characterization fusion vector or the preference fusion vector and each of the video vectors to be recommended, determine the recommendation value of each video vector to be recommended, wherein the video vector to be recommended is used to indicate Attribute information of the video to be recommended.

Specifically, the video attribute information of each video to be recommended is mapped to a vector representation. In this embodiment, the video attribute information is mapped and converted through the MLP network model, thereby outputting the corresponding video vector V to be recommended. According to the representation fusion vector or The dot product result of the preference fusion vector and each video vector to be recommended is used to realize two-tower modeling to calculate the recommended value of the video to be recommended, and the dot product result between the fusion vector and the video vector to be recommended is used as the video vector to be recommended The recommendation value of , that is, F V, combines the cross information of the long video viewing sequence and the short video viewing sequence and the attribute information of the video to be recommended to determine the rough sorting result of the video to be recommended; according to the preference fusion vector and the video vector to be recommended The dot product result between is used as the recommended value of the video vector to be recommended, that is, H·V, which combines the user's preference information for long video viewing sequences and short video viewing sequences and the attribute information of the video to be recommended to determine the video to be recommended The results of rough sorting.

In one embodiment, after converting the acquired attribute information of each video to be recommended into corresponding vectors of the video to be recommended, the method further includes:

The video vector to be recommended is added to the dot product result of the characterization fusion vector and the preference fusion vector to obtain the recommendation value of the video vector to be recommended.

Specifically, the vector of the video to be recommended is added to the result of the dot product of the characterization fusion vector and the preference fusion vector respectively, that is, F·V+H·V, so as to realize three-tower modeling calculation of the video to be recommended The recommendation value combines the cross information of the long video viewing sequence and the short video viewing sequence, the user's preference information for the long video viewing sequence and the short video viewing sequence, and the attribute information of the video to be recommended to determine the rough sorting result of the video to be recommended. In this way, the user's preference information for videos can be more fully and comprehensively captured, so as to provide a video rough sorting result more in line with the user's preferences.

Regardless of whether two-tower modeling or three-tower modeling is used, the modeling process also learns and trains the initial model according to the above-mentioned process for using the model, that is, performs cross-learning training on the sample sequence in the sample data set according to the above-mentioned process, so as to obtain A neural network model that implements the above process.

Fig. 2 is a schematic flowchart of a video recommendation method in an embodiment. It should be understood that although the various steps in the flow chart of FIG. 2 are displayed sequentially as indicated by the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least some of the steps in FIG. 2 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution of these sub-steps or stages The order is not necessarily performed sequentially, but may be performed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.

In one embodiment, as shown in Figure 3, a video recommendation device is provided, including:

An acquisition module 310, configured to acquire a video viewing set corresponding to the target user identifier, wherein the video viewing set includes at least two video viewing sequences with different video durations;

A fusion module 320, configured to extract corresponding feature vectors based on each of the video viewing sequences, and fuse each of the feature vectors to form a fusion vector;

The determination module 330 is configured to determine the recommended value of each video vector to be recommended according to the acquired inner product result of each video vector to be recommended and the fusion vector, wherein the video vector to be recommended is used to indicate the video vector to be recommended Attribute information of the video;

The push module 340 is configured to push the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommendation value.

In one embodiment, the acquiring module 310 is specifically configured to:

Obtaining a historical video collection corresponding to the target user identifier, wherein the historical video collection includes at least two historical video sequences with different video durations, and the historical video sequences include a plurality of video attribute vectors;

Performing one-hot encoding processing on each of the historical video sequences to obtain a corresponding video encoding sequence, wherein the video encoding sequence includes a plurality of one-hot encoding vectors corresponding to the video attribute vectors;

Perform dimensionality reduction processing on each of the video coding sequences to obtain the corresponding video viewing sequence, wherein the video viewing sequence includes a plurality of video description vectors.

In one embodiment, the acquiring module 310 is specifically configured to:

Multiplying each of the video encoding sequences by the first mapping matrix respectively to obtain the corresponding first dimensionality reduction sequence, wherein the first mapping matrix includes matrix parameters corresponding to video attributes;

Multiplying each of the video coding sequences by the second mapping matrix respectively to obtain the corresponding second dimensionality reduction sequence, wherein the second mapping matrix includes matrix parameters corresponding to the preference attributes.

In one embodiment, the fusion module 320 is specifically used for:

respectively performing mean pooling processing on each of the first dimensionality reduction sequences to obtain the corresponding characterization vectors;

Determine a primary fusion vector according to the dot product results between each of the characterization vectors;

performing fusion processing on the first-level fusion vector and all the first dimensionality reduction sequences to obtain a second-level fusion vector;

Adding the first-level fusion vector and the second-level fusion vector to form the representation fusion vector.

In one embodiment, the fusion module 320 is specifically used for:

Based on the association relationship between different video description vectors in each of the second dimensionality reduction sequences, determine a relationship learning vector corresponding to each of the video description vectors, wherein the relationship learning vector includes a target description vector and the target description vector The association relationship with each of the video description vectors in the second dimensionality reduction sequence, the target description vector is any one of the video description vectors in the second dimensionality reduction sequence;

determining a preference vector corresponding to each of the second dimensionality reduction sequences according to the confidence that the plurality of relational learning vectors corresponding to the same second dimensionality reduction sequence belong to each preference feature;

The preference vectors corresponding to each of the second dimensionality reduction sequences are fused to form the preference fusion vector.

In one embodiment, the determining module 330 is specifically configured to:

converting the acquired attribute information of each video to be recommended into a corresponding vector of the video to be recommended;

According to the inner product result between the characterization fusion vector or the preference fusion vector and each of the video vectors to be recommended, determine the recommendation value of each video vector to be recommended, wherein the video vector to be recommended is used to indicate Attribute information of the video to be recommended.

In one embodiment, the determining module 330 is specifically configured to:

The video vector to be recommended is added to the dot product result of the characterization fusion vector and the preference fusion vector to obtain the recommendation value of the video vector to be recommended.

Figure 4 shows a diagram of the internal structure of a computer device in one embodiment. Specifically, the computer device may be the server 120 in FIG. 1 . As shown in FIG. 4 , the computer equipment includes a processor, a memory, a network interface, an input device, and a display screen connected through a system bus. Wherein, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system, and may also store a computer program. When the computer program is executed by the processor, the processor may implement the video recommendation method. A computer program may also be stored in the internal memory, and when the computer program is executed by the processor, the processor may execute the video recommendation method. The display screen of the computer equipment may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment may be a touch layer covered on the display screen, or a button, a trackball or a touch pad provided on the casing of the computer equipment, or It can be an external keyboard, touchpad or mouse.

Those skilled in the art can understand that the structure shown in Figure 4 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation to the computer equipment on which the solution of the application is applied. The specific computer equipment can be More or fewer components than shown in the figures may be included, or some components may be combined, or have a different arrangement of components.

In one embodiment, the video recommendation apparatus provided in the present application may be implemented in the form of a computer program, and the computer program may run on the computer device as shown in FIG. 4 . Various program modules constituting the video recommendation apparatus can be stored in the memory of the computer equipment, for example, the acquisition module 310, the fusion module 320, the determination module 330 and the push module 340 shown in FIG. 3 . The computer program constituted by each program module enables the processor to execute the steps in the video recommendation method of each embodiment of the application described in this specification.

The computer device shown in FIG. 4 can execute the acquisition module 310 in the video recommendation device as shown in FIG. 3 to acquire the video viewing set corresponding to the target user identification, wherein the video viewing set includes at least two videos with different video durations. Watch the sequence. The computer device may use the fusion module 320 to extract corresponding feature vectors based on each of the video viewing sequences, and fuse each of the feature vectors to form a fusion vector. The computer device may determine the recommended value of each video vector to be recommended according to the obtained inner product result of each video vector to be recommended and the fusion vector through the determination module 330, wherein the video vector to be recommended is used to indicate Attribute information of the video to be recommended. The computer device may use the push module 340 to push the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommendation value.

In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and operable on the processor. method.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the method described in any one of the above-mentioned embodiments is implemented.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be realized by instructing related hardware through computer programs, and the programs can be stored in a non-volatile computer-readable storage medium When the program is executed, it may include the processes of the embodiments of the above-mentioned methods. Wherein, any references to memory, storage, database or other media used in the various embodiments provided in the present application may include non-volatile and/or volatile memory. Nonvolatile memory can include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Chain Synchlink DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

It should be noted that in this article, relative terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply these No such actual relationship or order exists between entities or operations. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

The above descriptions are only specific embodiments of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Accordingly, the present invention will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims (10)

1. A video recommendation method, characterized in that the method comprises: Obtain a video viewing set corresponding to the target user identifier, wherein the video viewing set includes at least two video viewing sequences with different video durations; Extracting corresponding feature vectors based on each of the video viewing sequences, and fusing each of the feature vectors to form a fusion vector; According to the obtained inner product result of each video vector to be recommended and the fusion vector, the recommendation value of each video vector to be recommended is determined, wherein the video vector to be recommended is used to indicate the attribute information of the video to be recommended; Pushing the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommendation value. 2. The method according to claim 1, wherein said acquiring the video viewing set corresponding to the target user identifier comprises: Obtaining a historical video collection corresponding to the target user identifier, wherein the historical video collection includes at least two historical video sequences with different video durations, and the historical video sequences include a plurality of video attribute vectors; Performing one-hot encoding processing on each of the historical video sequences to obtain a corresponding video encoding sequence, wherein the video encoding sequence includes a plurality of one-hot encoding vectors corresponding to the video attribute vectors; Perform dimensionality reduction processing on each of the video coding sequences to obtain the corresponding video viewing sequence, wherein the video viewing sequence includes a plurality of video description vectors. 3. The method according to claim 1, wherein the video viewing sequence comprises a first dimensionality reduction sequence and a second dimensionality reduction sequence, and each of the video coding sequences is subjected to dimensionality reduction processing to obtain corresponding The video viewing sequence includes at least one of the following: Multiplying each of the video encoding sequences by the first mapping matrix respectively to obtain the corresponding first dimensionality reduction sequence, wherein the first mapping matrix includes matrix parameters corresponding to video attributes; Multiplying each of the video coding sequences by the second mapping matrix respectively to obtain the corresponding second dimensionality reduction sequence, wherein the second mapping matrix includes matrix parameters corresponding to the preference attributes. 4. The method according to claim 3, wherein the feature vector comprises a characterization vector, the fusion vector comprises a characterization fusion vector, and the corresponding feature vectors are respectively extracted based on each of the video viewing sequences, and Merge each of the feature vectors to form a fusion vector, including: respectively performing mean pooling processing on each of the first dimensionality reduction sequences to obtain the corresponding characterization vectors; Determine a primary fusion vector according to the dot product results between each of the characterization vectors; performing fusion processing on the first-level fusion vector and all the first dimensionality reduction sequences to obtain a second-level fusion vector; Adding the first-level fusion vector and the second-level fusion vector to form the representation fusion vector. 5. The method according to claim 4, wherein the feature vector also includes a preference vector, the fusion vector includes a preference fusion vector, and the corresponding feature vectors are respectively extracted based on each of the video viewing sequences, And each of the feature vectors is fused to form a fusion vector, including: Based on the association relationship between different video description vectors in each of the second dimensionality reduction sequences, determine a relationship learning vector corresponding to each of the video description vectors, wherein the relationship learning vector includes a target description vector and the target description vector The association relationship with each of the video description vectors in the second dimensionality reduction sequence, the target description vector is any one of the video description vectors in the second dimensionality reduction sequence; determining a preference vector corresponding to each of the second dimensionality reduction sequences according to the confidence that the plurality of relational learning vectors corresponding to the same second dimensionality reduction sequence belong to each preference feature; The preference vectors corresponding to each of the second dimensionality reduction sequences are fused to form the preference fusion vector. 6. The method according to claim 5, wherein the inner product result of each obtained video vector to be recommended and the fusion vector is determined to determine the recommended value of each video vector to be recommended, including: converting the acquired attribute information of each video to be recommended into a corresponding vector of the video to be recommended; According to the inner product result between the characterization fusion vector or the preference fusion vector and each of the video vectors to be recommended, determine the recommendation value of each video vector to be recommended, wherein the video vector to be recommended is used to indicate Attribute information of the video to be recommended. 7. The method according to claim 6, wherein, after converting the acquired attribute information of each video to be recommended into the corresponding video vector to be recommended, the method further comprises: The video vector to be recommended is added to the dot product result of the characterization fusion vector and the preference fusion vector to obtain the recommendation value of the video vector to be recommended. 8. A video recommendation device, characterized in that the device comprises: An acquisition module, configured to acquire a video viewing set corresponding to the target user identifier, wherein the video viewing set includes at least two video viewing sequences with different video durations; A fusion module, configured to extract corresponding feature vectors based on each of the video viewing sequences, and fuse each of the feature vectors to form a fusion vector; A determination module, configured to determine the recommended value of each video vector to be recommended according to the acquired inner product result of each video vector to be recommended and the fusion vector, wherein the video vector to be recommended is used to indicate the video to be recommended attribute information; The push module is configured to push the video data corresponding to the video vector to be recommended to the terminal corresponding to the target user identifier in descending order of the recommendation value. 9. A computer device, comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, characterized in that, when the processor executes the computer program, any one of claims 1 to 7 is realized. A step of said method. 10. A computer-readable storage medium, on which a computer program is stored, characterized in that, when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images