Inferring Social Network Structure using Mobile Phone Data

We analyze 330,000 hours of continuous behavioral data logged by the mobile phones of 94 subjects, and compare these observations with selfreport relational data. The information from these two data sources is overlapping but distinct, and the accuracy of self-report data is considerably affected by such factors as the recency and salience of particular interactions. We present a new method for precise measurements of large-scale human behavior based on contextualized proximity and communication data alone, and identify characteristic behavioral signatures of relationships that allowed us to accurately predict 95% of the reciprocated friendships in the study. Using these behavioral signatures we can predict, in turn, individual-level outcomes such as job satisfaction.