Inferring explicit and implicit social ties simultaneously in mobile social networks

Abstract

Social tie inferring is to determine the type of relations, which is a significant task in social network analysis. Mobile social networks based on records from mobile phones, contain both interaction and spatial data of users, which may help to infer network structure, mine routines and recognize social ties. In mobile social networks, social ties are categorized into two types: explicit and implicit. Implicit social ties exist in real life but seldom interactions can be observed, and more analysis is shown in Appendix A. Traditionally, social tie recognition researches only focus on inferring explicit ties. But discovering implicit social ties is also meaningful, as it is a supplement to explicit tie recognition. Its applications include friend suggestion, unseen links detection and recommendation systems. Many previous researches have proved that multi-task learning can improve the performance when tasks are related. So it is reasonable to infer explicit and implicit social ties simultaneously. We propose a community-based factor graph model to infer explicit and implicit ties simultaneously. We firstly give empirical analysis and observe community features in mobile social networks. Then, we propose a community factor graph (CFG), whose node layer and relation layer are for the edge level features, and community layer is for community features. Finally, we set up experiments on real data and the results show that our model has best performance.