CN-122002059-A - Video recommendation method, device, equipment, medium and product

CN122002059ACN 122002059 ACN122002059 ACN 122002059ACN-122002059-A

Abstract

The application discloses a video recommendation method, a device, equipment, a medium and a product, which can be applied to various scenes such as an electronic commerce platform, a social media service, an education platform and the like, wherein the method comprises the steps of obtaining a video list; the method comprises the steps of counting video playing time in a video list, obtaining a playing time threshold of each video, determining a target prediction model based on the playing time threshold of each video and a depth learning model, and predicting playing time data of a target user for the target video by using the target prediction model so as to recommend the video for the target user based on the playing time data. According to the video recommendation method and device, the video watching time of the audience is considered, the playing time threshold of the video is divided based on the video watching time of the audience, so that the prejudice caused by the physical time of the video is eliminated, the interestingness of the user to the video is better reflected, the target prediction model is further determined based on the playing time threshold of the video and the depth learning model, the accuracy of the target prediction model is improved, and therefore video recommendation is accurately carried out.

Inventors

HOU DONGJIE

Assignees

腾讯科技（北京）有限公司

Dates

Publication Date: 20260508
Application Date: 20241101

Claims (14)

1.A video recommendation method, comprising: Acquiring a video list; Counting the video playing time length in the video list to obtain a playing time length threshold value of each video, wherein the playing time length threshold value is used for representing the video playing time length of a preset user proportion; determining a target prediction model based on the playing time threshold value and the deep learning model of each video; Predicting playing time length data of a target user aiming at a target video by using the target prediction model so as to recommend the video for the target user based on the playing time length data, wherein the playing time length data is used for representing the time length data of the target user for playing the target video.
2. The method for recommending videos according to claim 1, wherein the step of counting the video playing time in the video list to obtain the playing time threshold of each video comprises the steps of: counting the video playing time length in the video list to obtain the playing time length distribution data of each video; Calculating a playing time length threshold of each video by utilizing a sliding time window, wherein the playing time length threshold comprises a first playing time length threshold and a second playing time length threshold, the first playing time length threshold is used for representing the time length of playing each video by a first preset audience proportion, the second playing time length threshold is used for representing the time length of playing each video by a second preset audience proportion, and the first preset audience proportion is larger than the second preset audience proportion.
3. The video recommendation method of claim 2, wherein determining a target prediction model based on the play duration threshold and the depth learning model for each video comprises: acquiring initial training data; combining the playing time threshold value of each video with the initial training data according to the video ID to obtain target training data; acquiring a target loss function; And learning the target training data and the target loss function through the deep learning model to obtain the target prediction model.
4. The video recommendation method of claim 3 wherein said combining said video playback time thresholds with said initial training data according to video ID to obtain target training data comprises: identifying video IDs corresponding to all initial samples in the initial training data; Adding a playing time length threshold of the video matched with the video ID corresponding to each initial sample in the initial training data to the corresponding initial sample to form each target sample; and summarizing the target samples to obtain the target training data.
5. The video recommendation method of claim 3, wherein said obtaining a target loss function comprises: calculating a first weight and a second weight of each target sample based on the target training data; and determining the target loss function based on the first weight of each target sample, the second weight and a preset loss function.
6. The video recommendation method of claim 5, wherein calculating a first weight and a second weight for each target sample based on the target training data comprises: Extracting a first playing time length threshold value, a second playing time length threshold value and a target playing time length in each target sample aiming at each target sample in the target training data, wherein the target playing time length is used for representing the playing time length of a video played by a user once; Calculating a first weight of each target sample based on a first playing time length threshold value, a second playing time length threshold value and a target playing time length in each target sample, wherein the first weight is used for representing the weight of interest of a user to the video; And carrying out normalization processing on the target playing time length to obtain a second weight, wherein the second weight is used for representing the weight of the playing time length of the video played by the user.
7. The video recommendation method of claim 6, wherein calculating the first weight for each target sample based on the first play duration threshold, the second play duration threshold, and the target play duration in each target sample comprises: calculating a difference value between the target playing time length and the first playing time length threshold value to obtain a first difference value; Calculating the difference value between the second playing time length threshold value and the first playing time length threshold value to obtain a second difference value; And calculating the quotient of the first difference value and the second difference value to obtain the first weight of each target sample.
8. The method of claim 6, wherein the normalizing the target playing duration to obtain the second weight includes: Acquiring the size of a sliding time window; And calculating the quotient of the target playing duration and the size of the sliding time window to obtain the second weight.
9. The video recommendation method of claim 5, wherein the determining the target loss function based on the first weight, the second weight, and a preset loss function for each target sample comprises: Carrying out weighted fusion on the first weight and the second weight of each target sample to obtain fusion weight of each target sample; Performing mathematical transformation on the fusion weights of the target samples to obtain target weights of the target samples; acquiring the preset loss function; Determining a loss function of each target sample based on the target weight of each target sample and the preset loss function; and summarizing the loss functions of the target samples to obtain the target loss functions.
10. The video recommendation method of claim 9, wherein said determining a loss function for each target sample based on a target weight for each target sample and said preset loss function comprises: And calculating the product of the target weight of each target sample and the preset loss function to obtain the loss function of each target sample.
11. A video recommendation device, comprising: the acquisition module is used for acquiring a video list; The statistics module is used for counting the video playing time in the video list to obtain a playing time threshold of each video, wherein the playing time threshold is used for representing the video playing time of a preset user proportion; The prediction model determining module is used for determining a target prediction model based on the playing time threshold value and the depth learning model of each video; The video recommendation module is used for predicting playing time length data of a target user for a target video by using the target prediction model so as to recommend the video for the target user based on the playing time length data, wherein the playing time length data is used for representing the time length of playing the target video by the target user.
12. A computer device comprising a memory, and one or more processors communicatively coupled to the memory; Stored in the memory are instructions executable by the one or more processors to cause the one or more processors to implement the video recommendation method of any one of claims 1 to 10.
13. A computer readable storage medium comprising a program or instructions which, when run on a computer, implements the video recommendation method of any one of claims 1 to 10.
14. A computer program product comprising a computer program which, when executed by a processor, implements the video recommendation method of any one of claims 1 to 10.

Description

Video recommendation method, device, equipment, medium and product Technical Field The application relates to the technical field of computers, in particular to a video recommendation method, a video recommendation device, video recommendation equipment, video recommendation media and video recommendation products. Background Leisure and recreation by watching videos has become one of the main leisure modes of people. When a user watches a video, in order to improve the watching experience of the user, an operator often recommends the video to the user. At present, when video recommendation is performed to a user, the video recommendation is often performed by discretizing the physical duration distribution of the video, fitting an independent viewing duration prediction model (i.e., a playing duration prediction model) for each physical duration group, and using the obtained viewing duration prediction model. However, when the number of the physical duration groups increases in the method, the number of samples of each group is reduced, so that the statistics is reduced, the prediction performance of the viewing duration prediction model is further affected, and the accuracy of video recommendation is reduced. Disclosure of Invention The main purpose of the application is to provide a video recommendation method, a device, equipment, a medium and a product, which can embody the interest degree of a user on a video, and improve the accuracy of a target prediction model so as to accurately recommend the video. In order to achieve the above object, in a first aspect, the present application provides a video recommendation method, including: Acquiring a video list; counting the video playing time length in the video list, and obtaining a playing time length threshold value of each video, wherein the playing time length threshold value is used for representing the video playing time length of a preset user proportion; Determining a target prediction model based on a play duration threshold value and a depth learning model of each video; Predicting playing time length data of the target user aiming at the target video by using the target prediction model to recommend the video for the target user based on the playing time length data, wherein the playing time length data is used for representing the time length data of the target user for playing the target video. In an embodiment, counting video playing time in a video list to obtain a playing time threshold of each video includes: counting the video playing time length in the video list to obtain the playing time length distribution data of each video; Calculating a playing time length threshold of each video by utilizing the sliding time window, wherein the playing time length threshold comprises a first playing time length threshold and a second playing time length threshold, the first playing time length threshold is used for representing the time length of playing each video by a first preset audience proportion, the second playing time length threshold is used for representing the time length of playing each video by a second preset audience proportion, and the first preset audience proportion is larger than the second preset audience proportion. In one embodiment, determining a target prediction model based on a play duration threshold and a depth learning model for each video includes: acquiring initial training data; combining the playing time threshold value of each video with the initial training data according to the video ID to obtain target training data; acquiring a target loss function; and learning the target training data and the target loss function through the deep learning model to obtain a target prediction model. In one embodiment, combining the playing time threshold of each video with the initial training data according to the video ID to obtain the target training data includes: Identifying video IDs corresponding to all initial samples in the initial training data; Adding a playing time length threshold of the video matched with the video ID corresponding to each initial sample in the initial training data to the corresponding initial sample to form each target sample; And summarizing all the target samples to obtain target training data. In one embodiment, obtaining the target loss function includes: calculating a first weight and a second weight of each target sample based on the target training data; And determining a target loss function based on the first weight, the second weight and the preset loss function of each target sample. In one embodiment, calculating the first weight and the second weight for each target sample based on the target training data includes: extracting a first playing time length threshold value, a second playing time length threshold value and a target playing time length in each target sample aiming at each target sample in target training data, wherein the target playing time length is used for represe