Dual-level Hypergraph Representation Learning for Group Recommendation

Abstract

AbstractGroup Recommendation (GR) is the task of recommending items for a group of users. Most of existing studies adopt heuristic or attention-based preference aggregation strategies to learn group preferences, which ignores the composition of the group and suffers seriously from the problem of group-item interactions sparsity. In this paper, we propose a new group recommendation model based on Dual-level Hypergraph Representation Learning (called DHRL), which well models the group decision-making process by considering user-item interactions, group-item interactions and group-group interactions. Specifically, we design a member-level hypergraph convolutional network to learn group members’ personal preferences from user-item interactions. We also design a group-level hypergraph convolutional network to capture group preferences with full consideration of both group-item interactions and group-group interactions. Finally, we propose a joint training strategy to ease data sparsity by combining the group recommendation task with the user recommendation task. The experiments demonstrate the effectiveness and the efficiency of our proposed method compared to several state-of-the-art methods in terms of HR and NDCG.KeywordsGroup recommendationHypergraph convolutional networkGroup decision-makingRepresentation learningGroup-group interactions