Discovering Point-of-Interest Signatures Based on Group Features from Geo-social Networking Data

In recent years, location-based social networking services (LBSNSs) have become popular, generating a huge volume of geo-social networking data, such as check-in records and geo-tagged photos. The geo-social networking data provide a new source for discovering the real-world user behaviors. The information is useful for different applications, such as location prediction and point-of-interest (POI) recommendation. For LBSNSs, the research in POI recommendation have widely studied the user preferences over POIs and social influences between users. However, POIs are usually favored by or suitable for different kinds of groups, such as a small group, a tight group, or a close group. In this paper, we propose an approach to discovering POI signatures from geo-social networking data. For each POI, we first discover whether it has been visited by any groups of people and the features of these groups from user trajectories. We then generate the signature for each POI based on the discovered group features. We conduct experiments on the real data of the check-in records from Bright kite, and show the various kinds of POI signatures we found.