A Latent Group Model for Group Recommendation

Abstract

Increasingly, thousands of mobile services are provided by mobile Internet service portals. In order to push information that users may fond of, recommender system is needed. In some circumstances, recommendation to groups is necessary, e.g., Recommending movies to a group of friends. In reality, users are in some hidden social network, which can be viewed as groups. So group recommendation is proposed. Time efficiency is a key problem in mobile group recommendation. Research on group recommendation have concentrated on two approaches: aggregating members' ratings into a group profile and aggregating users' recommendations into a group recommendation list. This paper proposes a latent group model LGM, based on the assumption that users are influenced implicitly by some latent factors. LGM presents a novel route to detect groups by taking latent factors into account and makes users' profiles exist in latent factor format. Then users' latent factor profiles are aggregated into a group profile and multiplying method is used for group recommendation. This paper compares LGM with two approaches proposed before in efficiency and accuracy. It achieves better efficiency and accuracy for group recommendation on Movie Lens dataset.