Exploiting Implicit Social Relationship for Point-of-Interest Recommendation

The emergence of Location-based Social Network (LBSN) services allows users to share their check-ins, providing an excellent opportunity to build personalized Point-of-Interest (POI) recommender systems. Social network data which contains important context information has been demonstrated to have a significant effect on improving recommendation performances. However, explicit social relationships are usually partially available or even unavailable. The gap between the importance of social relationships and their partial availability or unavailability motivates us to study POI recommendation with implicit social relationships, which can well characterize users' preferences for POIs on both space and content. In this paper, we first extract implicit social relationships and estimate connection strengths by analyzing co-occurrences in both space and time with people's history check-in data. Then, we propose a new model named Implicit Social Relationship Enhanced POI Recommendation (ImSoRec) to incorporate implicit and explicit social relationships for POI recommendation. We conducted extensive experiments on two large-scale real-world location-based social networks datasets, and our experimental results show that our proposed ImSoRec model outperforms the state-of-the-art methods.