User Behavior Prediction And Recommendation Based On Con-Trastive Learning And Graph Learning

With the development of the Internet and information industry,a large number of news is disseminated through news app.In the process of communication,the news app platform will generate a large amount of data.The research and analysis of these data can make accurate recommendations,which can improve the economic benefits of the news app.This paper constructs a bipartite network based on the user click data of a news app,and uses the measurement metrics such as network density,degree distribution and cluster coefficient to explore the law of user click news.After generating the corresponding corrputed graph from the original bipartite graph through different data augment methods,the link prediction of the connecting edge in the bipartite network is carried out by using the combination of contrastive learning and graph neural network,the user behavior in the data set is analyzed,and subsequent recommendations are made.In the bipartite network constructed based on the data sampled by the news app,the network density is only 0.74%,and the degree distribution of news and user nodes approximately subject to the power-law distribution,and the interaction relationship is weak,which is in line with the distribution law of general large-scale interactive networks.Therefore,the data is suitable for prediction with the model used in this paper.By comparing different data augment methods during model training,this paper improves the AUC by 0.603%and the precision by 0.9741%.After fine-tuning the hyperparameters,the AUC is 85.8886%and the precision is 87.8705%.This paper further analyzes the prediction results.Based on the analysis of the prediction results of news nodes with different degrees,it is found that the precision of link prediction of news with different popularity is between 0.9 and 1.Therefore,most of the links predicted from the perspective of news are correct.We can find the corresponding users of potential connections from the perspective of news and carry out subsequent recommendation tasks.