DGC: Dynamic group behavior modeling that utilizes context information for group recommendation

In the real world, group recommendation that recommends items to a set of users (i.e., a group) is a challenging problem since it is very difficult to ensure the satisfaction of all group members with different preferences. Many existing group recommendation systems commonly use aggregation methods, which are insufficient to model group behavior where preference of a user as an individual is often changed when he/she is a member of a group. Some recent methods attempt to reflect this dynamic group behavior. However, they still have limitations in capturing complex relationships between items and group members. Moreover, previous approaches do not fully utilize useful context information available, and only use rating data. Reflecting context information together with ratings helps resolve a well-known data sparsity problem in group recommendation. In this paper, we propose a novel group recommendation framework called Dynamic Group behavior modeling that utilizes Context information for group recommendation(DGC). In DGC, we newly develop dynamic group behavior modeling that enables summarization of complex patterns in group decision making processes. To apply context information, firstly, we extract relevant context information from a heterogeneous information network (HIN) that contains rich information between various entities. Then, context information is properly applied to a group recommendation model by using semi-supervised learning that is composed of a supervised loss for label prediction and an unsupervised loss for context prediction. Experimental results show that our method provides significant performance improvement over other existing methods.