Social Network Discovery from Multiple Log Data through a Behavior Model

This paper proposes a new framework for inferring social networks of smart phone users from log data on web browsing and mobile application execution. In the framework, a behavior model is built by taking into account the effects of homophily, other users and adopted items. Social networks are then derived from the estimated behavior model by considering the semantics of the effects. Inference methods from multiple log data are also proposed in which behavior models are estimated under certain constraints by using an integrated effect. Experiments using three real log data of 130 smart phone users confirm the feasibility of the proposed framework.