Collaborative group-activity recommendation in location-based social networks

Location-based social networks (LBSNs) such as Foursquare, Google+ Local have become a popular platform for users to share their activities with family and friends. They provide rich information for us to study research issues of group recommendation services by exploiting the social and location characteristics of users and places. In this paper, we are interested in examining the effectiveness of modeling group dynamics for 'group recommendation' in LBSNs. We propose a novel hierarchical Bayesian model which jointly learns activities and group preferences by using topic models; and performs group recommendation using matrix factorization in a collaborative filtering framework. We show that our model allows for group preference learning by capturing location and user-group membership information and, it also handles data sparsity and cold start recommendation problems. A major advantage of our modeling framework is that we can interpret the learned group preferences using latent topics. Empirical experiments on a large LBSN dataset (Gowalla) shows that our model provides more effective group recommendation system than the state-of-the-art approaches. We show that the user preferences vary based on their groups, and users tend to exhibit a flair for novelty and exploration as part of a group. Our results reveal interesting insights into how the user and group preferences differ, and how the dominant user's behavior influences group's decisions.