Neural network collaborative filtering for group recommendation

Abstract

In the group recommender system, most of methods through aggregating individual preferences of each member in the group to group preference, which neglect the correlation among the members of the group. In this paper, group recommendation based on neural collaborative filtering (GNCF) and convolutional neural collaborative filtering (GCNCF) frameworks are proposed, which simulate the interaction between the members of the group and make recommendations directly for the group. GNCF and GCNCF frameworks predict group ratings by learning user-item interaction matrices. They project sparse vectors to dense vectors by utilizing the full connection layer, and improve the non-linear capability of the model by using the deep neural networks. Comparing with the traditional method, our method builds a new group recommendation model, and its effectiveness is well demonstrated through experiments.