Region partition using user mobility patterns based on topic model

Urban spatial structure has long been studied by geographers and economists to understand the development of cities using the activities extracted from surveys. With the popularity of mobile devices, massive urban sensing data has brought the opportunities to study human activities and city dynamics. Location based service has enormous growth in the recent years. Such abundant information benefits a variety of applications from urban planning to targeted advertising. In this paper, we perform an in-depth analysis of the correlations of the regions by conducting experiments with trajectories of millions of mobile phone subscribers. We present a method for the classification of cell regions from call data records generated by the pervasive cell phone network. By using latent model we have reduced the dimensions of regions spatial-temporal features and represent the functions of regions with probabilistic distribution. It is shown that our algorithm partitions the city into several regions which is well in line with the administrative area planning.