Hierarchical Fuzzy Graph Attention Network for Group Recommendation

Human's group activities have contributed to the development of group recommender systems. The group recommender system can provide personalised services for various online user groups through analysing groups' preferences. However, current group recommendation methods have failed to exploit complex relationships among users, groups and items when extracting groups' preferences. Meanwhile, most previous works are based on crisp techniques, which result in rigid preference profiling. Benefiting from the development of graph attention networks, this paper represents the complex relationships among users, groups and items as various graphs, including user-/group-item graph, user-group graph and user-user graph, and proposes a hierarchical fuzzy graph attention network (HGAT-F) to enhance fuzzy profiling for both groups and items. Experiments results on real world datasets show that HGAT-F has enhanced group recommendation than previous works.