Leveraging the fine-grained user preferences with graph neural networks for recommendation

With the explosion of information, recommendation systems have become important for users to find their interested information. Existing recommendation methods mainly utilize user historical interaction with items or user ratings to capture user past preferences. However, there is ignorance of various personalized reasons for users preferring an item, in which the reasons always dominate users’ preference strengths on the item. In addition, the linear nature of traditional recommendation methods makes them less effective in dealing with complex data. With the development of deep learning methods, graph neural networks provide an unprecedented opportunity for recommendations, since the user-item interactions can be naturally represented as a graph and the method can extract high-order complex relationships between users and items. In this paper, we propose a novel method leveraging the FIne-Grained user preferences with Graph Neural Networks (FigGNN) for recommendation to tackle these issues. More specifically, user-item interactions with user annotated tags and user ratings are constructed as a graph. In the process of graph message propagation, the user annotated tags are incorporated for understanding user preference reasons on items, and heterogeneous user rating levels are utilized for recognizing user preference strengths on items. Experiments have been conducted on the MovieLens dataset and the results show a superior performance of FigGNN over baselines in terms of precision and recall, which demonstrates the effectiveness of the proposed method.