Find in Library

User’s review text data and rating data, as two major information sources of the recommender system, reflect user’s preferences and item characteristics from two different perspectives. Many existing methods rely on one or the other data to make recommendations, ignoring the potential collaborative effects between the two. In terms of processing review data, the existing neural combination models mainly capture the local and continuous dependencies between adjacent words in the review text but have limited ability to capture the global and discontinuous dependencies. Therefore, we propose an enhanced recommendation model based on both the review text graph and the interaction graph. On the one hand, the model represents the review text of each user and item as a graph and uses the graph structure to capture the long-term, global, and discontinuous dependencies between words in the review text. A graph attention network based on connection relationships is used to aggregate the adjacency information of each node while taking word order relationships into account. On the other hand, the model builds an interaction graph based on user-item ratings for feature mining. The results of the two parts are combined to complete the prediction. We conduct experiments on three datasets and the results show that the proposed method can improve the recommendation performance.