Research on Neural Graph Collaborative Filtering Recommendation Model Fused With Item Temporal Sequence Relationships

Abstract

Graph neural network is blossoming recently,and it can explicitly express user-item high-order connectivity information,so it can significantly improve the recommendation performance when it applied to recommender system. However, the existing methods usually assume that the user’s interest is invariant,and there is insufficient explore to charactertize the user’s dynamic interest changes through the temporal sequences dependencies of items. In this paper, we propose a research on graph neural network collaborative recommendation model fused item temporal sequence relation, that is, a top-N hybrid recommendation model that fuses user-item interaction information and item temporal sequences dependencies. It divides the item temporal sequences into several groups of subsequences through the sliding window, constructs the item temporal sequences dependency graph, aggregates the characteristics of item temporal sequences information, and deeply depict the dynamic changes of users’ interests, and uses the bipartite graph neural network to map the high-dimensional information of user-item and item-item to the low-dimensional space. The hybrid embedding of user-item historical interaction information and item temporal sequences dependency information is realized, and the expression of user-item interaction sequence information is enhanced. Finally, through the user-item interaction graph and item sequence dependency graph, constructed a graph neural collaborative filtering recommendation framework including embedding layer,aggregation layer,propagation layer and prediction layer. Experiments demonstrate that the model performance has been significantly improved on datasets such as LastFM, Ciao and Douban.