Search

CN-115687690-B - Video recommendation method and device, electronic equipment and storage medium

CN115687690BCN 115687690 BCN115687690 BCN 115687690BCN-115687690-B

Abstract

The embodiment of the invention provides a video recommendation method, a device, electronic equipment and a storage medium, which relate to the technical field of network data processing and are used for acquiring attribute characteristics, positive behavior characteristics and negative behavior characteristics of a target user, inputting the attribute characteristics, the positive behavior characteristics and the negative behavior characteristics of the target user into a main network in a pre-trained recall model, inputting the negative behavior characteristics of the target user into a biasing network in the pre-trained recall model, fusing an output vector corresponding to the obtained main network and a biasing vector corresponding to the obtained biasing network to obtain interest expression vectors of the target user, calculating similarity of the interest expression vectors and expression vectors of a plurality of videos to be recommended, selecting N videos with highest similarity to recommend the target user, avoiding video content which is not interested by the recommended target user, and improving video recommendation precision.

Inventors

  • ZHU XIANWU
  • Jiang sanfeng

Assignees

  • 北京奇艺世纪科技有限公司

Dates

Publication Date
20260505
Application Date
20221009

Claims (9)

  1. 1. A video recommendation method, the method comprising: The method comprises the steps of obtaining attribute characteristics, positive behavior characteristics and negative behavior characteristics of a target user, wherein the positive behavior characteristics comprise characteristic information of a plurality of positive videos, the positive videos are videos with watching time length longer than a first preset time length of the target user, and the negative behavior characteristics comprise characteristic information of a plurality of negative videos, and the negative videos are videos with watching time length shorter than a second preset time length of the target user; Inputting the attribute characteristics, the positive behavior characteristics and the negative behavior characteristics of the target user into a main network in a pre-trained recall model; Vectorizing the attribute features, the feature information of the positive video and the feature information of the negative video to obtain attribute feature vectors, positive behavior feature vectors and negative behavior feature vectors; Predicting the weight of each forward video corresponding to the forward behavior feature according to the attribute feature vector and the forward behavior feature vector, and carrying out weighted fusion on the representation vector of each forward video through the weight of each forward video to obtain the vector representation of the forward behavior feature; Predicting the weight of each negative video corresponding to the negative behavior feature according to the attribute feature vector and the negative behavior feature vector, and carrying out weighted fusion on the representation vector of each negative video through the weight of each negative video to obtain the vector representation of the negative behavior feature; fusing the vector representation of the positive behavior characteristic and the vector representation of the negative behavior characteristic to obtain an output vector corresponding to the main network; inputting the negative behavior characteristics of the target user into a bias network in a pre-trained recall model to obtain a bias vector corresponding to the bias network; fusing the output vector corresponding to the main network and the bias vector corresponding to the bias network to obtain an interest expression vector of the target user; and calculating the similarity of the interest expression vector and the expression vectors of the videos to be recommended, and selecting N videos with the highest similarity to recommend to the target user.
  2. 2. The method of claim 1, wherein the inputting the negative behavioral characteristics of the target user into the biasing network in the pre-trained recall model to obtain the biasing vector corresponding to the biasing network comprises: inputting the negative behavior characteristics of the target user into a bias network in a pre-trained recall model; vectorizing the negative behavior characteristic through the bias network to obtain a negative behavior characteristic vector; splicing the negative behavior feature vector and the output vector corresponding to the main network; And outputting the spliced vector through a full connection layer of the offset network to obtain an offset vector corresponding to the offset network.
  3. 3. The method of claim 1, wherein the calculating the similarity between the interest representation vector and the representation vectors of the plurality of videos to be recommended, and selecting N videos with highest similarity to recommend to the target user, comprises: Calculating the similarity of the interest expression vector and the expression vectors of the plurality of videos to be recommended; Sequencing the plurality of videos to be recommended according to the sequence from big to small of the calculated similarity to obtain a sequenced video sequence; and selecting the first N videos in the video sequence to recommend the first N videos to the target user.
  4. 4. A recall model training method, the method comprising: Acquiring a positive sample video and a negative sample video, wherein the positive sample video is a video with a watching time length longer than a first preset time length of a plurality of target users, and the negative sample video is a video with a watching time length shorter than a second preset time length of the plurality of target users; Inputting the positive sample video and the negative sample video into a recall model to be trained to respectively obtain a positive sample video representation vector and a negative sample video representation vector; The method comprises the steps of acquiring and inputting attribute characteristics, positive behavior characteristics and negative behavior characteristics of a target user into a main network in a recall model to be trained, wherein the positive behavior characteristics comprise characteristic information of a plurality of positive videos, the positive videos are videos with watching time length longer than a first preset duration of the target user, the negative behavior characteristics comprise characteristic information of a plurality of negative videos, and the negative videos are videos with watching time length shorter than a second preset duration of the target user; Vectorizing the attribute features, the feature information of the positive video and the feature information of the negative video to obtain attribute feature vectors, positive behavior feature vectors and negative behavior feature vectors; Predicting the weight of each forward video corresponding to the forward behavior feature according to the attribute feature vector and the forward behavior feature vector, and carrying out weighted fusion on the representation vector of each forward video through the weight of each forward video to obtain the vector representation of the forward behavior feature; Predicting the weight of each negative video corresponding to the negative behavior feature according to the attribute feature vector and the negative behavior feature vector, and carrying out weighted fusion on the representation vector of each negative video through the weight of each negative video to obtain the vector representation of the negative behavior feature; fusing the vector representation of the positive behavior characteristic and the vector representation of the negative behavior characteristic to obtain an output vector corresponding to the main network; Inputting the negative behavior characteristics of the target user into a bias network in a recall model to be trained to obtain a bias vector corresponding to the bias network; Fusing the output vector corresponding to the main network and the bias vector corresponding to the bias network to obtain a predicted interest expression vector of the target user; calculating a first difference between the predicted interest representation vector and the positive sample video representation vector, and a second difference between the predicted interest representation vector and the negative sample video representation vector; And adjusting parameters of the recall model to be trained according to the first difference value and the second difference value, and returning to the step of inputting the positive sample video and the negative sample video into the recall model to be trained to respectively obtain a positive sample video representation vector and a negative sample video representation vector, and continuing training until a preset requirement is met to obtain the trained recall model.
  5. 5. A video recommendation device, the device comprising: The device comprises a feature acquisition module, a feature processing module and a feature processing module, wherein the feature acquisition module is used for acquiring attribute features, positive behavior features and negative behavior features of a target user, the positive behavior features comprise feature information of a plurality of positive videos, the positive videos are videos with watching time longer than a first preset time length of the target user, and the negative behavior features comprise feature information of a plurality of negative videos, and the negative videos are videos with watching time shorter than a second preset time length of the target user; The network output module is used for inputting the attribute characteristics, the positive behavior characteristics and the negative behavior characteristics of the target user into a main network in a pre-trained recall model, carrying out vectorization on the attribute characteristics, the characteristic information of the positive videos and the characteristic information of the negative videos to obtain attribute characteristic vectors, positive behavior characteristic vectors and negative behavior characteristic vectors, predicting the weights of all positive videos corresponding to the positive behavior characteristics according to the attribute characteristic vectors and the positive behavior characteristic vectors, carrying out weighted fusion on the representation vectors of all positive videos through the weights of all positive videos to obtain vector representations of the positive behavior characteristics, predicting the weights of all negative videos corresponding to the negative behavior characteristics according to the attribute characteristic vectors and the negative behavior characteristic vectors, and carrying out weighted fusion on the representation vectors of all negative behavior characteristics through the weights of all negative videos to obtain vector representations of the negative behavior characteristics; The vector fusion module is used for fusing the output vector corresponding to the main network and the offset vector corresponding to the offset network to obtain the interest expression vector of the target user; and the video recommendation module is used for calculating the similarity between the interest expression vector and the expression vectors of the plurality of videos to be recommended, and selecting N videos with the highest similarity to recommend the N videos to the target user.
  6. 6. The apparatus of claim 5, wherein the network output module comprises: the negative behavior characteristic input sub-module is used for inputting the negative behavior characteristic of the target user into the bias network in the pre-trained recall model; the negative behavior feature vectorization sub-module is used for vectorizing the negative behavior features through the bias network to obtain negative behavior feature vectors; The vector splicing sub-module is used for splicing the negative behavior characteristic vector and the output vector corresponding to the main network; and the offset vector generation sub-module is used for outputting the spliced vector through the full connection layer of the offset network to obtain the offset vector corresponding to the offset network.
  7. 7. A recall model training apparatus, the apparatus comprising: The system comprises a sample video acquisition module, a video processing module and a video processing module, wherein the sample video acquisition module is used for acquiring positive sample videos and negative sample videos, the positive sample videos are videos with watching time lengths longer than a first preset time length for a plurality of target users, and the negative sample videos are videos with watching time lengths shorter than a second preset time length for the plurality of target users; the sample input module is used for inputting the positive sample video and the negative sample video into a recall model to be trained to respectively obtain a positive sample video representation vector and a negative sample video representation vector; The vector prediction module is used for acquiring and inputting the attribute characteristics, the positive behavior characteristics and the negative behavior characteristics of the target user into a main network in a recall model to be trained; the forward behavior feature comprises feature information of a plurality of forward videos, wherein the forward videos are videos with watching time length longer than a first preset time length of the target user; the negative behavior feature comprises feature information of a plurality of negative videos, the negative videos are videos with watching time length smaller than a second preset time length, the attribute feature, the feature information of the positive videos and the feature information of the negative videos are subjected to vectorization to obtain attribute feature vectors, positive behavior feature vectors and negative behavior feature vectors, the weights of the positive videos corresponding to the positive behavior feature are predicted according to the attribute feature vectors and the positive behavior feature vectors, the representing vectors of the positive videos are subjected to weighted fusion through the weights of the positive videos to obtain vector representations of the positive behavior feature, the weights of the negative videos corresponding to the negative behavior feature are predicted according to the attribute feature vectors and the negative behavior feature vectors, the vector representations of the negative behavior feature vectors are obtained through weighted fusion of the weights of the negative videos, the vector representations of the positive behavior feature and the vector representations of the negative behavior feature vectors are fused to obtain output vectors corresponding to the main network, the output vectors corresponding to the main network are input to the bias network, the bias network is trained by the corresponding to the bias network, obtaining a predicted interest expression vector of the target user; A difference calculation module for calculating a first difference between the predicted interest representation vector and the positive sample video representation vector, and a second difference between the predicted interest representation vector and the negative sample video representation vector; and the parameter adjustment module is used for adjusting the parameters of the recall model to be trained according to the first difference value and the second difference value, returning the steps of inputting the positive sample video and the negative sample video into the recall model to be trained to obtain a positive sample video representation vector and a negative sample video representation vector respectively, and continuing training until the preset requirement is met to obtain a trained recall model.
  8. 8. The electronic equipment is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are communicated with each other through the communication bus; a memory for storing a computer program; a processor for carrying out the method steps of any one of claims 1-3 or 4 when executing a program stored on a memory.
  9. 9. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored therein a computer program which, when executed by a processor, implements the method steps of any of claims 1-3 or 4.

Description

Video recommendation method and device, electronic equipment and storage medium Technical Field The present invention relates to the field of network data processing technologies, and in particular, to a video recommendation method, a video recommendation device, an electronic device, and a storage medium. Background At present, with the rapid development of the internet, watching videos through various video platforms for leisure and relaxation is also a leisure mode for a plurality of people. In order to improve user experience, many video platforms often recommend videos possibly interesting to users, which not only facilitates users to watch, but also improves user experience. However, when a current video website recommends a video to a user, the video website often recommends the video according to a watching record of the user, and if the user watches a certain type of video, the video is recommended to the user. However, some videos that are not interesting to the user are often included in the viewing record of the user, resulting in a problem of low recommendation accuracy when making a recommendation according to the history. Disclosure of Invention The embodiment of the invention aims to provide a video recommendation method, a video recommendation device, electronic equipment and a storage medium, so as to avoid video content which is not interested by a recommendation target user and improve video recommendation accuracy. The specific technical scheme is as follows: according to a first aspect of an embodiment of the present invention, there is provided a video recommendation method, including: Acquiring attribute characteristics, positive behavior characteristics and negative behavior characteristics of a target user; Inputting the attribute characteristics, the positive behavior characteristics and the negative behavior characteristics of the target user into a main network in a pre-trained recall model to obtain an output vector corresponding to the main network; fusing the output vector corresponding to the main network and the bias vector corresponding to the bias network to obtain an interest expression vector of the target user; and calculating the similarity of the interest expression vector and the expression vectors of the videos to be recommended, and selecting N videos with the highest similarity to recommend to the target user. Optionally, the positive behavior feature comprises feature information of a plurality of positive videos, wherein the positive videos are videos with watching time length longer than a first preset time length of the target user, and the negative behavior feature comprises feature information of a plurality of negative videos, wherein the negative videos are videos with watching time length shorter than a second preset time length of the target user; Inputting the attribute characteristics, the positive behavior characteristics and the negative behavior characteristics of the target user into a main network in a pre-trained recall model to obtain an output vector corresponding to the main network, wherein the method comprises the following steps: Inputting the attribute characteristics, the positive behavior characteristics and the negative behavior characteristics of the target user into a main network in a pre-trained recall model; Vectorizing the attribute features, the feature information of the positive video and the feature information of the negative video to obtain attribute feature vectors, positive behavior feature vectors and negative behavior feature vectors; Predicting the weight of each forward video corresponding to the forward behavior feature according to the attribute feature vector and the forward behavior feature vector, and carrying out weighted fusion on the representation vector of each forward video through the weight of each forward video to obtain the vector representation of the forward behavior feature; Predicting the weight of each negative video corresponding to the negative behavior feature according to the attribute feature vector and the negative behavior feature vector, and carrying out weighted fusion on the representation vector of each negative video through the weight of each negative video to obtain the vector representation of the negative behavior feature; and fusing the vector representation of the positive behavior characteristic and the vector representation of the negative behavior characteristic to obtain an output vector corresponding to the main network. Optionally, the inputting the negative behavior characteristic of the target user into a bias network in a pre-trained recall model to obtain a bias vector corresponding to the bias network includes: inputting the negative behavior characteristics of the target user into a bias network in a pre-trained recall model; vectorizing the negative behavior characteristic through the bias network to obtain a negative behavior characteristic vector; splicing the negat