CN116955705A

CN116955705A - Video recommendation method and device

Info

Publication number: CN116955705A
Application number: CN202210416140.1A
Authority: CN
Inventors: 王步霖
Original assignee: Shanghai Bilibili Technology Co Ltd
Current assignee: Shanghai Bilibili Technology Co Ltd
Priority date: 2022-04-20
Filing date: 2022-04-20
Publication date: 2023-10-27

Abstract

The application provides a video recommendation method and a device, wherein the video recommendation method comprises the following steps: acquiring user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, wherein the viewing characteristics represent whether a plurality of historical viewing records of the target user are continuously and completely played; determining the recommendation weight of a first video to be recommended to the target user according to the user information, the video information of the first video to be recommended and the watching characteristics, wherein the first video to be recommended is any one of the plurality of videos to be recommended; and determining target videos from the videos to be recommended according to the recommendation weight of each video to be recommended for the target user, and recommending the target videos to the target user. The method can effectively improve video recommendation efficiency and accuracy.

Description

Video recommendation method and device

Technical Field

The application relates to the technical field of computers, in particular to a video recommendation method. The application also relates to a video recommendation device, a computing device and a computer readable storage medium.

Background

With the continuous development of computer technology, various video platforms are growing, and accurately recommending videos required by users to users has become an important means for improving video viewing quantity. When the user sees the recommended video which is the video required by the user, the user can watch the recommended video without thinking, so that the user is facilitated, and the development of a video platform is facilitated.

In the prior art, a recommended video is generated by acquiring a historical browsing record of a user and analyzing the requirement of the user. However, in the above method, since the video recorded in the history browsing record is not all the video that the user likes to watch, the recommendation system is easy to misjudge the preference of the user, so that the recommendation accuracy is reduced.

Disclosure of Invention

In view of this, the embodiment of the application provides a video recommendation method. The application also relates to a video recommending device, a computing device and a computer readable storage medium, which are used for solving the technical defect of low accuracy of video recommendation in the prior art.

According to a first aspect of an embodiment of the present application, there is provided a video recommendation method, including:

acquiring user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, wherein the viewing characteristics represent whether a plurality of historical viewing records of the target user are continuously and completely played;

Determining the recommendation weight of a first video to be recommended to the target user according to the user information, the video information of the first video to be recommended and the watching characteristics, wherein the first video to be recommended is any one of the plurality of videos to be recommended;

and determining target videos from the videos to be recommended according to the recommendation weight of each video to be recommended for the target user, and recommending the target videos to the target user.

According to a second aspect of an embodiment of the present application, there is provided a video recommendation apparatus including:

the system comprises a first acquisition module, a second acquisition module and a first display module, wherein the first acquisition module is configured to acquire user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, and the viewing characteristics characterize whether a plurality of historical viewing records of the target user are continuously and completely played;

a determining module configured to determine a recommendation weight of a first video to be recommended for the target user according to the user information, video information of the first video to be recommended and the viewing feature, wherein the first video to be recommended is any one of the plurality of videos to be recommended;

And the recommending module is configured to determine target videos from the videos to be recommended according to the recommending weight of each video to be recommended for the target user, and recommend the target videos to the target user.

According to a third aspect of embodiments of the present application, there is provided a computing device comprising a memory, a processor and computer instructions stored on the memory and executable on the processor, the processor implementing the steps of the video recommendation method when executing the computer instructions.

According to a fourth aspect of embodiments of the present application, there is provided a computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the video recommendation method.

According to the video recommendation method provided by the application, the user information of the target user, the video information of a plurality of videos to be recommended and the watching characteristics of the target user are obtained, wherein the watching characteristics represent whether a plurality of historical watching records of the target user are continuously and completely played; determining the recommendation weight of a first video to be recommended to the target user according to the user information, the video information of the first video to be recommended and the watching characteristics, wherein the first video to be recommended is any one of the plurality of videos to be recommended; and determining target videos from the videos to be recommended according to the recommendation weight of each video to be recommended for the target user, and recommending the target videos to the target user. Through the watching characteristics of the target user, user information and video information of each video to be recommended are corrected, so that the recommendation weight of each video to be recommended for the target user is determined, the accuracy of the recommendation weight is improved, the preference of the user can be judged more accurately, the favorite video is recommended to the user more accurately, namely, the accuracy of video recommendation is improved, and the user viscosity is further improved.

Drawings

FIG. 1 is a flowchart of a video recommendation method according to an embodiment of the present application;

FIG. 2 is a schematic view showing the effect of a history of viewing records in a video recommendation method according to an embodiment of the present application;

FIG. 3 is a process flow diagram of a video recommendation method according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of a video recommendation device according to an embodiment of the present application;

FIG. 5 is a block diagram of a computing device according to one embodiment of the application.

Detailed Description

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application may be embodied in many other forms than those herein described, and those skilled in the art will readily appreciate that the present application may be similarly embodied without departing from the spirit or essential characteristics thereof, and therefore the present application is not limited to the specific embodiments disclosed below.

The terminology used in the one or more embodiments of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of the application. As used in one or more embodiments of the application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used in one or more embodiments of the present application refers to any or all possible combinations of one or more of the associated listed items.

It should be understood that, although the terms first, second, etc. may be used in one or more embodiments of the application to describe various information, these information should not be limited by these terms. These terms are only used to distinguish one type of information from another. For example, a first may also be referred to as a second, and similarly, a second may also be referred to as a first, without departing from the scope of one or more embodiments of the application. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.

First, terms related to one or more embodiments of the present application will be explained.

Automatic simulcasting: in the video recommendation process, after the current playing video is finished, if the user has no other operation, the phenomenon that the next recommended video is automatically played can be started.

And (3) invalid playing: a video is playing, but the user is not watching, and this play behavior is not valid for the recommender system and is therefore called invalid play.

First, a video recommendation method provided by the present application will be briefly described.

The video recommended scene of the internet television is a video recommended scene which starts to be gradually enlarged in the recent years, and has a lot of differences compared with the traditional video recommended scenes such as web pages, mobile phones and the like. The existing technical schemes of video recommendation of internet televisions can be roughly divided into the following two main categories: the viewing history of the Internet television scene is simply combined, and related contents are recommended to the user; the ordering method under the traditional video recommendation scene is applied to the Internet television video recommendation scene, and information is obtained from the play history of the user through a machine learning method, so that the user is recommended to possibly like the content. The user requirements are analyzed by acquiring the historical browsing records of the user, and the recommended video is generated.

However, the above method does not consider the characteristic of automatic video simulcast, that is, the user does other things while watching the video, does not pause the current video, and automatically starts playing the next recommended video after the current video is played, so that a large number of videos which are not actually watched by the user appear in the play history of the user, and further, the recommendation system makes misjudgment on the preference of the user, thereby reducing the recommendation accuracy.

Therefore, the application provides a video recommendation method, which comprises the steps of obtaining user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, wherein the viewing characteristics represent whether a plurality of historical viewing records of the target user are continuously and completely played; determining the recommendation weight of a first video to be recommended to the target user according to the user information, the video information of the first video to be recommended and the watching characteristics, wherein the first video to be recommended is any one of the plurality of videos to be recommended; and determining target videos from the videos to be recommended according to the recommendation weight of each video to be recommended for the target user, and recommending the target videos to the target user. Through the watching characteristics of each historical watching record of the target user, user information and video information of each video to be recommended are corrected, so that the recommendation weight of each video to be recommended for the target user is determined, the accuracy of the recommendation weight is improved, the preference of the user can be judged more accurately, the favorite video is recommended to the user more accurately, namely, the accuracy of video recommendation is improved, and further, the user viscosity is improved.

In the present application, a video recommendation method is provided, and the present application relates to a video recommendation apparatus, a computing device, and a computer-readable storage medium, which are described in detail in the following embodiments one by one.

Fig. 1 shows a flowchart of a video recommendation method according to an embodiment of the present application, which specifically includes the following steps:

step 102: and acquiring user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, wherein the viewing characteristics represent whether a plurality of historical viewing records of the target user are continuously and completely played.

The execution subject for implementing the video recommendation method may be a computing device having a video recommendation function, such as a server, a terminal, etc. having a video recommendation function.

Specifically, the target user refers to a user who needs to recommend a video for the target user, such as a user who is ready to watch a video, and a user who is about to watch one video and is ready to watch the next video; the user information refers to information corresponding to a target user, such as video types, ages, sexes and the like that the user likes to watch; the video to be recommended is a candidate video recommended for the target user; the video information refers to attribute information of a video to be recommended, such as a video name, a video type, a video content profile, actors corresponding to the video, a video duration and the like; the history viewing record refers to a record of each time a user opens the video viewing platform to view the video, and the history viewing record can contain information such as playing time (for example, 9 points and 35 minutes viewing on day 1 and 20 months) of the played video, playing progress (for example, 60% of the played video), video covers, video names (for example, "certain grid"), video collection number (for example, first collection) and the like; the continuous and complete playing means at least one video which is continuously played and has a playing progress of 100 percent, or at least one video which is automatically simulcast; the viewing feature refers to a continuous complete playing mark corresponding to each played video in a plurality of historical viewing records of the target user, if 1 continuous complete playing video is before a third played video in a certain historical viewing record, the continuous complete playing mark corresponding to the third played video is 1, and if a fourth played video in a certain historical viewing record is a 4 th video continuous complete playing, the continuous complete playing mark corresponding to the fourth played video is 4.

In practical applications, there are various ways of acquiring the user information of the target user, the video information of the plurality of videos to be recommended and the viewing characteristics of the target user, for example, an operator may send a video recommendation instruction to an execution subject, or send an instruction for acquiring the user information of the target user, the video information of the plurality of videos to be recommended and the viewing characteristics of the target user, and accordingly, the execution subject starts to acquire the user information of the target user, the video information of the plurality of videos to be recommended and the viewing characteristics of the target user after receiving the instruction; the server may also automatically acquire user information of the target user, video information of a plurality of videos to be recommended, and viewing characteristics of the target user every preset time period, for example, after the preset time period, the server with the video recommendation function automatically acquires the user information of the target user, the video information of the plurality of videos to be recommended, and the viewing characteristics of the target user; or after a preset time length, the terminal with the video recommendation function automatically acquires the user information of the target user, the video information of a plurality of videos to be recommended and the video information of the watching characteristics of the target user, which are stored locally. The present specification does not limit any way to obtain the user information of the target user, the video information of the plurality of videos to be recommended, and the viewing characteristics of the target user.

In one or more optional embodiments of the present application, when obtaining user information of a target user, video information of a plurality of videos to be recommended, and viewing features of the target user, the user information corresponding to the user and the viewing features of the target user may be obtained first, and then the plurality of videos to be recommended may be recalled from the video library based on the user information and a preset recall algorithm.

Step 104: and determining the recommendation weight of the first video to be recommended to the target user according to the user information, the video information of the first video to be recommended and the watching characteristics, wherein the first video to be recommended is any one of the plurality of videos to be recommended.

On the basis of acquiring user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, further, recommendation weights of the videos to be recommended for the target user are respectively determined according to the user information, the video information and the viewing characteristics.

In particular, the recommendation weight characterizes a weight or value by which the video to be recommended may be recommended.

In practical application, after obtaining user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, whether each historical viewing record representing the target user is continuously and completely played or not needs to be subjected to viewing characteristics, and the user information and the video information of a first video to be recommended are processed, so that recommendation weight of the first video to be recommended for the target user is obtained. Traversing each video to be recommended to obtain the recommendation weight of each video to be recommended for the target user.

In one or more optional embodiments of the present application, in order to improve video recommendation efficiency, when determining a recommendation weight of a first video to be recommended for the target user according to the user information, video information of the first video to be recommended, and the viewing feature: the continuous and complete playing probability corresponding to the first video to be recommended can be determined based on the watching characteristics of the target user, which characterize whether each historical watching record is continuously and completely played or not, according to the video information of the first video to be recommended; and then determining the recommendation weight of the first video to be recommended for the target user according to the user information and the continuous complete play probability corresponding to the first video to be recommended.

In one or more alternative embodiments of the present application, in order to improve the accuracy of video recommendation, the video preference information of the target user may be determined first, and the recommendation weight may be determined based on the video preference information of the target user. That is, the determining the recommendation weight of the first video to be recommended for the target user according to the user information, the video information of the first video to be recommended, and the viewing feature may be implemented as follows:

Determining video preference information of the target user according to the user information and the viewing characteristics, wherein the video preference information characterizes the video information preferred to be watched by the target user;

and determining the recommendation weight of the first video to be recommended for the target user according to the video preference information and the video information of the first video to be recommended.

Specifically, the video preference information refers to video information that the target user likes to watch, such as a liked video type, a liked video duration, and the like.

In practical application, the video preference information of the target user can be determined based on the user information of the target user and the viewing characteristics of the target user, which characterize whether each historical viewing record is continuously and completely played, and then the recommendation weight of the first video to be recommended to the target user is calculated according to the first video to be recommended and the video preference information. Therefore, as the user information carries information such as the video types which the user likes to watch, and the watching characteristics for representing whether each historical watching record is continuously and completely played can also reflect the information such as the video types which the user likes to watch from a certain angle, the video preference information of the user is determined by combining the information and the information, the comprehensiveness of the video preference information can be ensured, the recommendation weight of the first video to be recommended for the user is determined by combining the video information of the first video to be recommended, the credibility and the accuracy of the recommendation weight can be improved, and the accuracy of video recommendation is further improved.

For example, the user information of the target user includes videos that like a certain star, and each history view record of the target user has a certain variety that is continuously and completely played, that is, the view feature is a certain variety that is continuously and completely played, then according to the user information and the view feature, the video preference information of the target user can be determined to be videos that like a certain star and/or a certain variety, then the association degree of the video preference information and the video information of the first video to be recommended is calculated, and further the recommendation weight of the first video to be recommended to the target user is determined according to the association degree.

Step 106: and determining target videos from the videos to be recommended according to the recommendation weight of each video to be recommended for the target user, and recommending the target videos to the target user.

On the basis of determining the recommendation weight of the first video to be recommended to the target user according to the user information, the video information of the first video to be recommended and the viewing characteristics of the plurality of historical viewing records, further, screening out the target video according to the recommendation weight of each video to be recommended to the target user and recommending the target video to the target user.

Specifically, the target video refers to a video to be recommended, which needs to be recommended to the user.

In practical application, after the recommendation weight of each video to be recommended for the recommendation weight of the target user is obtained, the target video is screened from the multiple videos to be recommended based on preset screening conditions and the recommendation weight corresponding to each video to be recommended. Further, the target video is recommended to the target user.

In one or more optional embodiments of the present application, when determining the target video in the videos to be recommended according to the recommendation weights, a threshold is set, and the videos to be recommended with recommendation weights greater than the threshold are determined as the target videos. After the target video is determined, the target video is recommended to the target user. Therefore, some high-quality target videos with high relevance to the target user can be recommended to the target user, and video recommendation efficiency and accuracy are improved.

In one or more alternative embodiments of the present application, in order to improve video recommendation efficiency, the recommendation weights may be ranked, and then a target video in each video to be recommended may be determined according to the ranking result. That is, the determining, according to the recommendation weight of each video to be recommended for the target user, the target video from the plurality of videos to be recommended includes:

And arranging the recommendation weights of each video to be recommended for the target users from large to small, and determining the video to be recommended corresponding to the recommendation weights of N before arrangement as the target video, wherein N is a positive integer.

In practical application, for example, recommendation weights of videos to be recommended for target users can be arranged from large to small, and videos to be recommended corresponding to the recommendation weights of N before arrangement are determined as target videos, wherein N is a preset numerical value; or sequencing the videos to be recommended according to the sequence of the recommendation weights from large to small, and determining the N videos to be recommended before sequencing as target videos. Thus, when the number of target videos recommended to the user is limited, the video recommendation efficiency can be improved.

In addition, the recommendation weights of the videos to be recommended for the target users can be arranged from small to large, and the videos to be recommended corresponding to the last M recommendation weights are determined to be the target videos, wherein M is a positive integer; or sequencing the videos to be recommended according to the sequence of the recommendation weights from small to large, and determining the last M videos to be recommended as target videos.

In one or more optional embodiments of the present application, before determining the recommendation weight of each video to be recommended for the target user, a pre-trained video recommendation model may be further obtained, then user information, video information of each video to be recommended, and viewing characteristics of the target user are input into the video recommendation model, and the video recommendation model processes the user information, the video information of each video to be recommended, and the viewing characteristics of the target user, so as to obtain the recommendation weight corresponding to each video to be recommended. That is, before determining the recommendation weight of the first video to be recommended for the target user according to the user information, the video information of the first video to be recommended, and the viewing feature, the method further includes:

The method comprises the steps of obtaining a pre-trained video recommendation model, wherein the video recommendation model is obtained based on sample video training carrying first identification information and second identification information, the first identification information represents continuous complete playing marks corresponding to the sample video, and the second identification information represents whether the sample video is effectively played or not;

accordingly, the determining the recommendation weight of the first video to be recommended for the target user according to the user information, the video information of the first video to be recommended and the viewing feature includes:

and inputting the user information, the video information of the first video to be recommended and the watching characteristics into the video recommendation model, and determining the recommendation weight of the first video to be recommended for the target user.

Specifically, the object recommendation model refers to a pre-trained neural network model, such as a neural network model and a probabilistic neural network model; the first identification information refers to a continuous and complete playing mark corresponding to a sample video, if there are 5 continuous and complete playing marks corresponding to a sample video before the sample video in a certain history viewing record, the continuous and complete playing mark corresponding to the sample video is 5, and if the sample video in the certain history viewing record is a 6 th video continuously and completely played, the continuous and complete playing mark corresponding to a fourth played video is 6.

In practical application, after user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user are obtained, a video recommendation model which is trained in advance based on sample videos carrying first identification information and second identification information is obtained, wherein the first identification information represents continuous and complete playing marks corresponding to the sample videos, and the second identification information represents whether the sample videos are effectively played. And then inputting the user information, the video information of the first video to be recommended and the watching characteristics of the target user into a video recommendation model, analyzing and processing the user information, the video information of the first video to be recommended and the watching characteristics of the target user by the video recommendation model, and outputting the recommendation weight of the first video to be recommended for the target user. The user information, the video information of the first video to be recommended and the watching characteristics of the target user are processed through the pre-trained video recommendation model, so that the speed and accuracy of acquiring the recommendation weight can be improved.

Before the pre-trained video recommendation model is obtained, the neural network model needs to be trained so as to obtain the video recommendation model with the recommendation weight determining function. That is, before the pre-trained video recommendation model is obtained, the method further includes:

Acquiring a sample video set and a preset neural network model, wherein the sample video set comprises a plurality of sample videos carrying first identification information and second identification information;

extracting video information of any sample video from the sample video set, and inputting the video information of the sample video into the neural network model to obtain a prediction result corresponding to the sample video;

calculating a loss value according to the prediction result, the first identification information and the second identification information carried by the sample video;

and according to the loss value, adjusting model parameters of the neural network model, continuously executing the step of extracting video information of any sample video from the sample video set, and determining the trained neural network model as a video recommendation model under the condition that a preset training stop condition is reached.

Specifically, the neural network model refers to a certain neural network model, such as a convolutional neural network model and a probabilistic neural network model, which are specified in advance; the sample video set refers to training samples of the neural network model; the prediction result refers to the output of a preset neural network model; the training stop condition may be that the loss value is smaller than or equal to a preset threshold, or that the number of iterative training reaches a preset iteration value, or that the loss value converges, that is, the loss value is not reduced as training is continued.

In practical applications, there are various ways of acquiring the sample video set and the neural network model, for example, an operator may send a training instruction of the neural network model to an execution subject, or send an instruction of acquiring the sample video set and the neural network model, and correspondingly, the execution subject starts to acquire the sample video set and the neural network model after receiving the instruction; the server may automatically acquire the sample video set and the neural network model every preset time, for example, after the preset time, the server with the model training function automatically acquires the sample video set and the neural network model; or after a preset time length, the terminal with the model training function automatically acquires a local sample video set and a neural network model. The present description does not set any limitation on the manner in which the sample video set and neural network model are obtained.

After a sample video set containing a plurality of sample videos carrying first identification information and second identification information and a preset neural network model are obtained, training the neural network model based on the sample video set to obtain a video recommendation model: the method comprises the steps of selecting video information of one sample video from a sample video set, inputting the video information of the sample video into a preset neural network model, and performing analysis, calculation and other processes on the video information of the sample video by the neural network model to obtain a prediction result corresponding to the sample video. Further, calculating a loss value according to a prediction result, first identification information and second identification information carried by the sample video, adjusting model parameters of a neural network model according to the loss value under the condition that a preset training stop condition is not met, and then selecting video information of one sample video from a sample video set again to perform the next training; and under the condition that the preset training stopping condition is reached, determining the trained neural network model as a video recommendation model. Therefore, the sample video carrying the first identification information and the second identification information is used for training the model to be trained, the accuracy and the rate of determining the recommendation weight of the video to be recommended by the video recommendation model can be improved, and the robustness of the video recommendation model is improved.

In one or more alternative embodiments of the present application, when calculating the loss value, the prediction result, the first identification information and the second identification information may be directly input into a preset loss function to obtain the loss, or the initial loss value may be determined according to the prediction result and the second identification information, and the loss value is calculated based on the initial loss value and the first identification information. Namely, the calculating a loss value according to the prediction result, the first identification information and the second identification information carried by the sample video includes:

calculating an initial loss value according to the prediction result and the second identification information carried by the sample video;

determining a gradient coefficient corresponding to the sample video according to first identification information carried by the sample video;

and calculating a loss value according to the gradient coefficient and the initial loss value.

Specifically, the initial loss value refers to a loss value obtained by comparing the predicted value with the tag value, that is, comparing the predicted result with the second identification information; the gradient coefficient characterizes the probability that the sample video is not played inefficiently, the smaller the gradient coefficient, the more likely the sample video is played inefficiently.

In practical application, the prediction result and the second identification information carried by the sample video can be compared or input into a preset first loss function, such as a cross entropy loss function, an L1 loss function and the like, so as to obtain an initial loss value; further, the first identification information carried by the sample video is input into a preset gradient coefficient determining function, and the gradient coefficient corresponding to the sample video is determined. And finally, inputting the gradient coefficient and the initial loss value into a preset second loss function to obtain a loss value. Therefore, the accuracy of the loss value can be improved, the neural network model can be converged rapidly, and the accuracy of the video recommendation model is further improved.

For example, the preset gradient coefficient determining function may be shown in equation 1, and the second loss function may be shown in equation 2.

λ=1/log (n+5) -0.5 (formula 1)

In equations 1 and 2, N represents first identification information, λ represents a gradient coefficient, w represents a model parameter, J (w) represents a first loss function, and L represents a loss value. I.e. the loss value is the gradient coefficient multiplied by the partial derivative of the model parameter by the first loss function.

It should be noted that, when determining the gradient coefficient corresponding to the sample video according to the first identification information carried by the sample video, if the continuous complete playing mark accords with a preset mark condition, determining that the gradient coefficient corresponding to the sample video information is 1; if the continuous complete playing mark does not meet the preset mark condition, inputting the continuous complete playing mark into a preset gradient coefficient determining function, and determining a gradient coefficient corresponding to the sample video information. In addition, the video recommendation model is a deep neural network model trained by using historical viewing records and is trained by a gradient descent method.

In one or more alternative embodiments of the application, adjustment values for the model parameters may be determined from the loss values, and then the model parameters are adjusted based on the adjustment values. That is, the adjusting the model parameters of the neural network model according to the loss value includes:

Inputting an initial value of the model parameter and the loss value into a preset parameter determination model aiming at any model parameter in the neural network model, and determining an adjustment value of the model parameter;

and adjusting the model parameters in the neural network model according to the adjustment values.

Specifically, the neural network model has at least one model parameter; the initial value is a value corresponding to a model parameter before being adjusted; the adjustment value is a corresponding value after adjustment of a certain model parameter.

In practical application, after obtaining the loss value, model parameters in the neural network model need to be adjusted one by one, and any one of the model parameters is taken as an example for explanation: and determining an adjustment value of the model parameter according to the initial value and the loss value of the model parameter, namely inputting the initial value and the loss value of the model parameter into a parameter determination model to obtain the adjustment value of the model parameter. The values of the model parameters are then adjusted from the initial values to the adjusted values. Therefore, the adjustment value of the model parameters can be rapidly determined, the adjustment rate of the model parameters is improved, and further the model training efficiency is improved.

The parameter determination model is shown in equation 3 or equation 4, for example.

w _j :＝w _j -alpha L (3)

3 and 4,w _j For the J-th model parameter, α is the learning rate, λ represents the gradient coefficient, J (w _j ) Representing a first loss function, L andindicating the loss value. Wherein equation 4 represents that the adjustment value of the jth model parameter is equal to the initial value of the jth model parameter, minus the learning rate (α) times the gradient coefficient (λ) times the partial derivative of the first loss function with respect to the jth model parameter. In the training process, the influence of gradients generated by the play records which are possibly invalid to the neural network model is reduced, so that a more accurate video recommendation model is finally obtained.

Therefore, aiming at the problem of automatic video simulcast, special correction is carried out on the sample video, a video recommendation model with better recommendation effect is produced through a gradient descent method, and more accurate recommendation effect is realized.

In one or more optional embodiments of the present application, the sorted sample video set may be directly obtained, and multiple history viewing records of any user may also be obtained, that is, the played video in the multiple history viewing records may be obtained, to determine the sample video set. That is, the acquiring the sample video set includes:

acquiring a plurality of historical viewing records of any user, wherein the historical viewing records contain video information of at least one played video;

For any historical viewing record, determining first identification information of each played video according to the playing progress of each played video in the historical viewing record;

acquiring second identification information of each played video;

and marking each played video by using the first identification information and the second identification information of each played video to obtain a sample video set.

Specifically, the played video refers to a video in a history of viewing; the playing progress refers to the ratio of the playing end node of the played video to the video duration, for example, the playing end node of the played video is 10 minutes, that is, the played video is closed at the playing node for 10 minutes, and if the video duration of the played video is 20 minutes, the playing progress is 50%.

In practical applications, a plurality of historical viewing records of a user are obtained, and each historical viewing record comprises video information of one or more played videos. Taking one of the history viewing records as an example, the following description will be given: and acquiring the playing progress of each played video in the historical viewing record, and then determining the first identification information of each played video according to the playing progress. And then obtaining second identification information of each played video, and marking the first identification information and the second identification information on each played video to obtain a plurality of sample videos carrying the first identification information and the second identification information. Therefore, the sample video is determined according to the first identification information and the second identification information, the confidence and reliability of the sample video are improved, and the neural network model is trained based on the sample video, so that the model training time can be shortened, and the robustness of the video recommendation model is improved.

It should be noted that, the first identification information and the second identification information need to be in one-to-one correspondence with each played video. Labeling the played video A according to the first identification information and the second identification information of the played video A; and labeling the played video B according to the first identification information and the second identification information of the played video B.

In one or more alternative embodiments of the present application, the second identification information of each played video may be determined based on a statistical number of played videos in each historical viewing record and a first number of continuous complete plays of the played videos in each historical viewing record. Namely, the specific implementation process of obtaining the second identification information of each played video may be as follows:

determining the statistical quantity of played videos in each historical viewing record and the first quantity of continuously and completely played videos in each historical viewing record;

judging whether the first quantity corresponding to any historical viewing record is larger than the statistical quantity or not according to any historical viewing record;

if yes, determining that second identification information of each played video in the historical viewing record is an invalid playing identification;

If not, determining the second identification information of each played video in the historical viewing record as an effective playing identification.

Specifically, the statistical quantity may be an average value of played videos in each history viewing record, may be a median of played videos in each history viewing record, or may be a minimum value of played videos in each history viewing record, which is not limited in the present application; the first number refers to the number of played videos continuously and completely played in a certain historical viewing record; the effective playing identifier characterizes that the played video is effective playing; the invalid playback identifier characterizes the played video as invalid playback.

In practical application, the playing quantity of the played videos in each historical viewing record is obtained, and then the statistical quantity is determined according to each playing quantity. And simultaneously, determining the first quantity of continuous and complete playing of the played videos in each historical viewing record according to the playing progress of the played videos in each historical viewing record. Further, it is determined whether the first number corresponding to each of the history viewing records is greater than the statistical number: the second identification information of each played video in the historical viewing records with the first number being greater than the statistical number is an invalid playing identification; and the second identification information of each played video in the historical viewing records with the first number smaller than or equal to the statistical number is an effective playing identification. Thus, the played video which is probably invalid to play can be discovered from the historical watching record of the user, if the user watches the video normally, the interaction behavior can be generated with high probability, if the interaction behavior (including praise, coin-in, collection, pause and the like) is not performed continuously for a long time, but a plurality of videos are continuously and completely played, and the invalid playing record is high in probability. Therefore, the accuracy of the second identification information can be improved to a certain extent.

For example, the number K (K > 0) of videos watched by the user at each time of opening the video playing application is obtained according to the history of watching the large disc, namely the statistical number is K. If there are M recordings in a certain user's history of viewing, M is the first number of consecutive full plays of video. If M > K, the user is likely not actually watching the video in the period of time, records the M records, and determines the second identification information of each played video in the historical watching record as an invalid playing identification. And if M is less than or equal to K, determining the second identification information of each played video in the historical viewing record as an effective playing identification.

In addition, after the second identification information is determined, the M videos need to be recorded, and marked to indicate that N videos have been continuously and completely played before each video, where N is a natural number.

In one or more alternative embodiments of the present application, the first identification information, i.e., the number of videos that are continuously and completely played before the continuously and completely played is marked as the played video. That is, the determining the first identification information of each played video according to the playing progress of each played video in the historical viewing record may be implemented as follows:

Determining a second number of specified playing videos corresponding to the played videos to be continuously and completely played according to any played video in the historical viewing record, wherein the specified playing videos are played videos which are played before the played videos in the historical viewing record;

and determining the second number as the first identification information of the played video.

Specifically, the second number refers to the number of videos that are continuously and completely played before a certain played video in a certain historical viewing record.

In practical application, for any played video in the history viewing record, counting the number of continuous and complete plays of the played video played before the played video in the history viewing record, namely, the second number; the second number is then used as the first identification information for the played video.

For example, in a history viewing record, there are three played videos, where the first played video is not continuously and completely played, the second played video is continuously and completely played, and the first played video is continuously and completely played, then the first identification information of the first played video is 0, the first identification information of the second played video is 0, and the first identification information of the third played video is 1.

In addition, only the first identification information of the played video that is continuously and completely played may be marked, and the first identification information of the played video that is not continuously and completely played may be defaulted to 0.

Referring to fig. 2, fig. 2 is a schematic diagram showing an effect of a history viewing record in a video recommendation method according to an embodiment of the present application: wherein, video 1, video 2 and video 7 are not completely played, and video 3, video 4, video 5 and video 6 are completely played, i.e. 4 videos continuously and completely played appear in the history viewing record. Assuming the statistics k=2.3, these 4 videos are marked:

a. before the video 3, no continuous complete play exists, and the first identification information is 0;

b. 1 video is continuously and completely played before the video 4, and the first identification information is 1;

c. 2 videos are continuously and completely played before the video 5, and the first identification information is 2;

d. video 6 was preceded by 3 consecutive plays of video in full, the first identification information being 3.

The video recommendation method provided by the application comprises the steps of obtaining user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, wherein the viewing characteristics represent whether a plurality of historical viewing records of the target user are continuously and completely played; determining the recommendation weight of a first video to be recommended to the target user according to the user information, the video information of the first video to be recommended and the watching characteristics, wherein the first video to be recommended is any one of the plurality of videos to be recommended; and determining target videos from the videos to be recommended according to the recommendation weight of each video to be recommended for the target user, and recommending the target videos to the target user. Through the watching characteristics of the target user, user information and video information of each video to be recommended are corrected, so that the recommendation weight of each video to be recommended for the target user is determined, the accuracy of the recommendation weight is improved, the preference of the user can be judged more accurately, the favorite video is recommended to the user more accurately, namely, the accuracy of video recommendation is improved, and the user viscosity is further improved.

The application of the video recommendation method provided by the application in an internet television scene is taken as an example in the following description with reference to fig. 3, and the video recommendation method is further described. Fig. 3 shows a processing flow chart of a video recommendation method according to an embodiment of the present application, which specifically includes the following steps:

step 302: and acquiring a plurality of historical viewing records of any user, wherein the historical viewing records contain video information of at least one played video.

Step 304: for any historical viewing record, determining first identification information of each played video according to the playing progress of each played video in the historical viewing record.

Optionally, determining the first identification information of each played video according to the playing progress of each played video in the historical viewing record includes:

the second number is determined as the first identification information of the played video.

Step 306: a statistical number of played videos in each historical viewing record is determined, and a first number of played videos in each historical viewing record are continuously and completely played.

Step 308: and judging whether the first quantity corresponding to any historical viewing record is larger than the statistical quantity or not according to any historical viewing record.

If yes, go to step 310, if no, go to step 312.

Step 310: and determining the second identification information of each played video in the historical viewing record as an invalid playing identification.

Step 312: and determining the second identification information of each played video in the historical viewing record as a valid playing identification.

Step 314: and marking each played video by using the first identification information and the second identification information of each played video to obtain a sample video set.

Step 316: the method comprises the steps of obtaining a preset neural network model, extracting video information of any sample video from a sample video set, and inputting the video information of the sample video into the neural network model to obtain a prediction result corresponding to the sample video.

Step 318: and calculating an initial loss value according to the prediction result and the second identification information carried by the sample video.

Step 320: and determining a gradient coefficient corresponding to the sample video according to the first identification information carried by the sample video.

Step 322: the loss value is calculated from the gradient coefficient and the initial loss value.

Step 324: for any model parameter in the neural network model, inputting an initial value and a loss value of the model parameter into a preset parameter determination model, and determining an adjustment value of the model parameter.

Step 326: and adjusting the model parameters in the neural network model according to the adjustment values.

Step 328: and continuing to execute the step of extracting video information of any sample video from the sample video set, and determining the trained neural network model as a video recommendation model under the condition that the preset training stop condition is reached.

Step 330: and acquiring user information of the target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, wherein the viewing characteristics represent whether a plurality of historical viewing records of the target user are continuously and completely played.

Step 332: and inputting the user information, the video information of the first video to be recommended and the watching characteristics into a video recommendation model, and determining the recommendation weight of each video to be recommended for the target user, wherein the first video to be recommended is any one of a plurality of videos to be recommended.

Optionally, the video recommendation model determines video preference information of the target user based on the feature information of whether each historical viewing record of the target user is continuously and completely played according to the user information; and determining the recommendation weight of each video to be recommended for the target user according to the video preference information and the video information of each video to be recommended.

Step 334: and arranging the recommendation weights of each video to be recommended for the target users from large to small, determining the video to be recommended corresponding to the recommendation weights of the N before arrangement as the target video, and recommending the target video to the target users, wherein N is a positive integer.

The application provides a video recommendation method, which corrects user information and video information of each video to be recommended by representing whether each historical watching record is continuously and completely played or not by a target user, so as to determine the recommendation weight of each video to be recommended for the target user, improve the accuracy of the recommendation weight, and more accurately judge the preference of the user, thereby more accurately recommending favorite videos to the user, namely improving the accuracy of video recommendation and further improving the viscosity of the user.

Corresponding to the method embodiment, the application also provides an embodiment of the video recommending apparatus, and fig. 4 shows a schematic structural diagram of the video recommending apparatus according to an embodiment of the application. As shown in fig. 4, the apparatus includes:

a first obtaining module 402, configured to obtain user information of a target user, video information of a plurality of videos to be recommended, and viewing features of the target user, where the viewing features characterize whether a plurality of historical viewing records of the target user are continuously and completely played;

A determining module 404 configured to determine a recommendation weight of a first video to be recommended for the target user according to the user information, video information of the first video to be recommended, and the viewing feature, wherein the first video to be recommended is any one of the plurality of videos to be recommended;

and a recommending module 406 configured to determine a target video from the videos to be recommended according to the recommending weight of each video to be recommended for the target user, and recommend the target video to the target user.

In one or more alternative embodiments of the application, the determining module 404 is configured to:

In one or more alternative embodiments of the present application, the apparatus further includes a second acquisition module configured to:

Accordingly, the determining module 404 is further configured to:

In one or more alternative embodiments of the application, the apparatus further comprises a training module configured to:

In one or more alternative embodiments of the application, the training module is further configured to:

Acquiring second identification information of each played video;

In one or more alternative embodiments of the present application, the recommendation module 406 is further configured to:

The application provides a video recommendation device, which is used for acquiring user information of a target user, video information of a plurality of videos to be recommended and viewing characteristics of the target user, wherein the viewing characteristics represent whether a plurality of historical viewing records of the target user are continuously and completely played; determining the recommendation weight of a first video to be recommended to the target user according to the user information, the video information of the first video to be recommended and the watching characteristics, wherein the first video to be recommended is any one of the plurality of videos to be recommended; and determining target videos from the videos to be recommended according to the recommendation weight of each video to be recommended for the target user, and recommending the target videos to the target user. User information and video information of videos to be recommended are corrected through viewing characteristics of each historical viewing record of the target user, so that recommendation weight of each video to be recommended for the target user is determined, accuracy of the recommendation weight is improved, preference of the user can be judged more accurately, favorite videos are recommended to the user more accurately, namely accuracy of video recommendation is improved, and user viscosity is further improved.

The above is an exemplary scheme of a video recommendation apparatus of the present embodiment. It should be noted that, the technical solution of the video recommendation device and the technical solution of the video recommendation method belong to the same concept, and details of the technical solution of the video recommendation device, which are not described in detail, can be referred to the description of the technical solution of the video recommendation method.

Fig. 5 illustrates a block diagram of a computing device 500, provided in accordance with an embodiment of the present application. The components of the computing device 500 include, but are not limited to, a memory 510 and a processor 520. Processor 520 is coupled to memory 510 via bus 530 and database 560 is used to hold data.

Computing device 500 also includes access device 540, access device 540 enabling computing device 500 to communicate via one or more networks 560. Examples of such networks include public switched telephone networks (PSTN, public Switched Telephone Network), local Area Networks (LAN), wide area networks (WAN, wideAreaNetwork), personal area networks (PAN, personalAreaNetwork), or combinations of communication networks such as the internet. The access device 540 may include one or more of any type of network interface, wired or wireless, such as a network interface card (NIC, network Interface Controller), such as an IEEE802.11 wireless local area network (WLAN, wireless LocalAreaNetwork) wireless interface, a worldwide interoperability for microwave access (Wi-MAX, worldwide Interoperability forMicrowave Acess) interface, an ethernet interface, a universal serial bus (USB, universal Serial Bus) interface, a cellular network interface, a bluetooth interface, a Near Field Communication (NFC) interface, and so forth.

In one embodiment of the application, the above-described components of computing device 500, as well as other components not shown in FIG. 5, may also be connected to each other, such as by a bus. It should be understood that the block diagram of the computing device illustrated in FIG. 5 is for exemplary purposes only and is not intended to limit the scope of the present application. Those skilled in the art may add or replace other components as desired.

Computing device 500 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., tablet, personal digital assistant, laptop, notebook, netbook, etc.), mobile phone (e.g., smart phone), wearable computing device (e.g., smart watch, smart glasses, etc.), or other type of mobile device, or a stationary computing device such as a desktop computer or PC. Computing device 500 may also be a mobile or stationary server.

Wherein the processor 520, when executing the computer instructions, implements the steps of the video recommendation method.

The foregoing is a schematic illustration of a computing device of this embodiment. It should be noted that, the technical solution of the computing device and the technical solution of the video recommendation method belong to the same concept, and details of the technical solution of the computing device, which are not described in detail, can be referred to the description of the technical solution of the video recommendation method.

An embodiment of the application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of the video recommendation method as described above.

The above is an exemplary version of a computer-readable storage medium of the present embodiment. It should be noted that, the technical solution of the storage medium and the technical solution of the video recommendation method belong to the same concept, and details of the technical solution of the storage medium, which are not described in detail, can be referred to the description of the technical solution of the video recommendation method.

The foregoing describes certain embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The computer instructions include computer program code that may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, randomAccess Memory), an electrical carrier signal, a telecommunication signal, a software distribution medium, and so forth.

It should be noted that, for the sake of simplicity of description, the foregoing method embodiments are all expressed as a series of combinations of actions, but it should be understood by those skilled in the art that the present application is not limited by the order of actions described, as some steps may be performed in other order or simultaneously in accordance with the present application. Further, those skilled in the art will appreciate that the embodiments described in the specification are all preferred embodiments, and that the acts and modules referred to are not necessarily all required for the present application.

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.

The preferred embodiments of the application disclosed above are intended only to assist in the explanation of the application. Alternative embodiments are not intended to be exhaustive or to limit the application to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the application and the practical application, to thereby enable others skilled in the art to best understand and utilize the application. The application is limited only by the claims and the full scope and equivalents thereof.

Claims

1. A video recommendation method, comprising:

2. The method of claim 1, wherein the determining the recommendation weight for the first video to be recommended for the target user based on the user information, the video information of the first video to be recommended, and the viewing characteristics comprises:

3. The method of claim 1, wherein prior to determining the recommendation weight for the first video to be recommended for the target user based on the user information, video information for the first video to be recommended, and the viewing characteristics, further comprising:

4. The method of claim 3, wherein prior to the obtaining the pre-trained video recommendation model, further comprising:

5. The method of claim 4, wherein calculating a loss value based on the prediction result, the first identification information and the second identification information carried by the sample video, comprises:

6. The method of claim 4, wherein adjusting model parameters of the neural network model based on the loss values comprises:

7. The method of any of claims 4-6, wherein the acquiring a sample video set comprises:

Acquiring second identification information of each played video;

8. The method of claim 7, wherein the obtaining the second identification information of each played video comprises:

9. The method of claim 7, wherein determining the first identification information of each of the played videos according to the playing progress of each of the played videos of the historical viewing record comprises:

10. The method of claim 1, wherein the determining the target video from the plurality of videos to be recommended according to the recommendation weight of each video to be recommended for the target user comprises:

11. A video recommendation device, comprising:

12. A computing device comprising a memory, a processor, and computer instructions stored on the memory and executable on the processor, wherein the processor, when executing the computer instructions, performs the steps of the method of any one of claims 1-10.

13. A computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the method of any one of claims 1-10.