Knowledge graph-based graph neural network models for multi-perspective modeling of group preferences

Abstract

The purpose of group recommendation is to recommend items that all users in a group may like; therefore, the modeling target of group recommendation is not individual preference features, but group preference features, which are very complex in the context of online social platforms. In this paper, in order to better model group preferences, We propose the model Knowledge graph-based Multi-view Attention Group Recommendation (KMAGR) to model the group preference relationship in three aspects: 1) adding knowledge mapping relationships as side information to the model; 2) using attention mechanisms and graph neural network structures to model group purchase intentions; 3) in addition to modeling group preferences from the user’s perspective, we use group-user and group-intent multiple perspectives to model group preferences. We conducted experiments on two real online social datasets, and the experimental results proved that KMAGR outperformed other state-of-the-art models in group recommendation. Adding knowledge graph information and identifying group intent to the group recommendation system can greatly improve the effectiveness of group recommendation, while the critical path aggregation mechanism improves the explainability of recommendation results.