The Method Of Session-based Recommendation With Graph Neural Networks

With the popularity of intelligent terminals,the generation and dissemination of information become more and more convenient,which leads to "information overload".People can take advantage of the recommendation system to filter information and find appropriate information from a large amount of information.So far,recommendation system has not only become an indispensable tool in enterprises,but also become a hot research field.In many of the latest real scenes,there are many cases where historical operation data and user configuration cannot be obtained.Therefore,the recommendation system cannot continuously track and record the behavior data of non-login users,resulting in the poor effect of the existing recommendation system.The session-based recommendation system can effectively resolve this problem,which can model and analyze the limited behavior sequence data of anonymous users,and predict the next recommendation content.Based on graph neural network,this paper conducts an in-depth study of session-based recommender systems.Aiming at the problem that the graph neural network cannot express the long-term interest dependence of the session well,a Multi-layer Attention Graph Neural Network for Session-based Recommendation(MLA-GNN)method is proposed,which can effectively capture the information propagation in the session sequence and the complex item conversion relationship in the session,and the attention mechanism is used to adaptively assign weights to the items in the current session,so as to better express the long-term global interest of the session.On the basis of the MLA-GNN method,in view of the lack of context information in the graph and the inability to guarantee the uniqueness of the relationship matrix,a method based on the context enhancement of the graph(GCEC)is proposed.It expresses the context in the graph accurately and reasonably,and can give the nodes importance weights and ensure the uniqueness of the coding.Aiming at the problem that the lack of position information in the self-attention mechanism makes the model unable to distinguish the importance of items in different positions,a method(APE)of the attention mechanism integrating position information is proposed to improve the discrimination ability.Finally,combining GCEC and APE,a Graph Context Enhanced Coding and Attention Position Enhanced GNN for Session-based Recommendation method(GCEC-APE-GNN)is proposed.The above two methods are compared with other methods on the real datasets,and the ablation experiment analysis and parameter visualization analysis are carried out,which verifies the rationality and effectiveness of the methods proposed in this paper.