Discovering and Predicting User Routines by Differential Analysis of Social Network Traces

The study of human activity patterns traditionally relies on the continuous tracking of user location. Although accurate in pinpointing meaningful locations and often adequate for predicting user movements in the near-term, these methods fail to account for the reasons of the presence of a user in a given location. We approach the problem of activity pattern discovery from a new perspective: instead of attempting to infer "user routines" by collecting sensor data of increasing volume and scope, we explore the participatory sensing potential of mobile social networks, on which users voluntarily disclose information on their location and the venues they are attending. In this paper, we present automated techniques for filtering, aggregating, and processing social networking traces to extract descriptions of user routines. We report our findings about user routines based on a data set of public Foursquare check-ins and geo-tagged Twitter messages collected from 825 users around the Tokyo metropolitan area over twelve months. We then attach meaningful labels to the features we observe in the routines based on available metadata, and evaluate the potential of user routines for predicting future location and activity of a user.