Multi-Interest Aware Graph Convolution Network for Social Recommendation

Zhengyi Guo, Yanmin Zhu, Zhaobo Wang, Mengyuan Jing

Abstract: Social recommendation endeavors to harness users’ social connections to enhance recommender systems. Graph Neural Network (GNN) has gained traction for its robust capacity in managing graph data. However, previous social recommendation works have failed to fully consider the crucial role of user distinct interests, which hinders their ability to accurately model complex user preferences and negatively affected the modeling of social influence. To tackle this challenge, we introduce a multi-interest approach to GNN-based social recommendation. Specifically, we firstly utilize a dynamic routing algorithm to cluster user interests from the items they have interacted with. Subsequently, we use a similarity-weighted GCN operation to capture user relationships within the social network. Finally, we use the aggregate the multiple interest representations for prediction. Our comprehensive experiments underscore the consistent superiority of our model compared with state-of-the-art competitors on real-world datasets.