Learning Sequential Mobility and User Preference for new Location Recommendation in Online Social Networks

The fast expansion during the recent years of online social networks, such as Twitter, Facebook, or Foursquare, is making available an enormous and continuous stream of user-generated contents including information on human mobility within urban context. In particular, online social networks allows for the collection of geo-tagged data obtained through the GPS readings of phones through which users have the possibility to tag posts, photos and videos with geographical coordinates. In this context, recommending the future position of a mobile object is key for the implementations of several applications aiming at improving mobility within urban areas. The paper proposes a location recommendation approach that exploits geo-tagged data on social networks. The approach integrates user preference, sequential mobility and geographic constraints. The recommendation task is formulated as a similarity problem among the visiting and mobility profiles of users, accounting the mobility sequentiality in the patterns. Two ranking metrics are introduced to predict places the user could like. The metrics are then combined into an overall recommendation ranking function. The candidate locations are then ranked according to the two similarity measures. The experimental results obtained by using a real-world dataset of tweets show that the proposed method is effective in recommending unseen locations, outperforming representative state-of-the-art approaches.