Dong Li

Group recommendation aims to recommend relevant items that cater to the preferences of a group of users. Its quality is greatly influenced by the conformity psychology exhibited by both individual users and the group as a whole. However, current techniques often overlook the subtleties and dynamics of conformity. To enhance the accuracy and interpretability of group recommendation, we propose a novel self-supervised Conformity-aware Group Recommendation model named ConfGR, which leverages multi-motif HyperGraph Convolution Networks (HGCN) to seamlessly integrate two crucial perspectives of conformity: social selection and social influence. Specifically, we first design social-motif HGCN and retail-motif HGCN respectively to dynamically capture the changes in members’ conformity within different groups. Then we develop groupon-motif HGCN and a cross-group line graph to collaboratively learn groups’ conformity. Finally, we integrate self-supervised learning into the training of our model to jointly enhance group embeddings under different motifs by finding the best trade-off between members’ conformity and groups’ conformity. We evaluate our model on three real-world datasets and the experimental results show the superiority of our model compared to state-of-the-art group recommendation models.