Inferring Implicit Social Ties in Mobile Social Networks

Mobile social networks provide a platform to collect users' mobility information via mobile devices, and foster many location-aware services, which enable us to infer mobile users' implicit social ties from their mobility information. Some researches have focused on identifying social ties from users' cooccurrences. However, they fail to consider that users meet at different times and places indicates different social tie strength. Furthermore, statistics show that most users with social ties meet with each other few times and even never meet in geography, making social tie inference more challenging. In this paper, we propose a unified framework to infer implicit social ties. Specifically, we explore different aspects of social ties on causing users' co-occurrences, such as location popularity, co-occurrence diversity, and users' mobility behaviors, and further incorporate them to capture the spatial-temporal dynamics for accurately distinguishing social ties from coincidences. For the case of few or no co-occurrences, we build three types of networks: location network, user-location cross network and incomplete user network. Then, a mutual impact factor graph model is proposed to infer missing social ties in the incomplete user network by transferring knowledge extracted from given location and user-location cross network. Experiments conducted on datasets from a real mobile social network show not only the superiority of distinguishing social ties from coincidences but also validate the predictability of social ties although mobile users have few interactions in the physical world.