Creating Social Network Models from Sensor Data

Complex macro-social phenomena can arise from simple micro-level behavior without any global coordination (e.g. racial segregation in neighborhoods can occur simply from individuals wanting to avoid being in the minority even in a population that prefers diversity [6]). Such simple local rules and assumptions are only rarely empirically grounded because they require studying a given social system at an individual scale and at a spatial and temporal resolution not possible using manual non-automated techniques. As a result, few studies have attempted to analyze social interactions that involve micro-level behavior. We are interested in grounding simple behavioral assumptions with real observations, and in testing the effects of these behaviors at the macro-level dynamics of groups and larger social systems. To this end, we instrumented a subject population with wearable sensors that recorded them as they went about their lives over the course of 9 months. We believe this longitudinal dataset will offer fertile opportunities to explore the type of questions described above. By continuously studying individuals in a social system, our goal is to discover the simple rules people follow or the behaviors they exhibit and how these affect the structure and dynamics of social networks. For example, is the interaction likelihood of two people proportional to their physical proximity, or instead to the stylistic similarity of their conversational dynamics [1] or body language? How predictive are the social network topologies that will emerge?