Multi-View Interactive Compromise Learning for Group Recommendation

Abstract

Due to the prevalence of group activities in daily life, there is an urgent need for group-based recommendation generation, known as group recommendation tasks. The final choice of a group involves the compromise and consensus among its members, which drives the decision-making process. However, existing methods for aggregating personal information lack consideration for the macro-level aspects of the group and fail to capture the compromises that occur in group interactions. To address these limitations, we propose the Multi-View Interactive Compromise Learning (MICL) approach. Specifically, MICL introduces the group-view adaptive graph transformer, member-view hypergraph aggregation network, and item-view tripartite graph augmentation to capture the interactive compromises among group members from group-view, member-view, and item-view, ultimately enabling to learn multi-view interactive compromise from both the group-item and user-item interactions. Extensive experiments demonstrate significant advantages of our proposed MICL over existing state-of-the-art solutions, and ablation studies confirm the robustness and efficiency.