Characterizing and modeling people movement from mobile phone sensing traces
With the ubiquity of mobile phones, a high accuracy of characterizing and modeling people movement is achievable. The knowledge about people's mobility enables many applications including highly efficient planning of cities' resources and network infrastructures, or dissemination of safety alerts. However, characterizing and modeling people movement remain very challenging due to difficulties in (a) capturing, cleaning, analyzing and storing real traces, and (b) achieving accurate predictions of different future contexts. In this paper, we present our effort in measuring and capturing phone sensory data as real traces, cleaning up measurements, and constructing prediction models. Specifically, we discuss design methodology, learned lessons from the implementation and deployment of a large-scale scanning system on 123 Google Android phones for 6 months at University of Illinois campus. We also conduct a characterization study on collected traces and present new findings in location visit pattern, location popularity, and contact pattern. Finally, we exploit joint location/contact traces to derive: (1) predictive models of missing contacts, and (2) prediction framework that provides future contextual information of people movement including locations, stay duration, and social contacts.