Extracting Semantics of Individual Places from Movement Data by Analyzing Temporal Patterns of Visits

Abstract

Data reflecting movements of people, such as GPS or GSM tracks, can be a source of information about mobility behaviors and activities of people. Such information is required for various kinds of spatial planning in the public and business sectors. Movement data by themselves are semantically poor. Meaningful information can be derived by means of interactive visual analysis performed by a human expert; however, this is only possible for data about a small number of people. We suggest an approach that allows scaling to large datasets reflecting movements of numerous people. It includes extracting stops, clustering them for identifying personal places of interest (POIs), and creating temporal signatures of the POIs characterizing the temporal distribution of the stops with respect to the daily and weekly time cycles and the time line. The analyst can give meanings to selected POIs based on their temporal signatures (i.e., classify them as home, work, etc.), and then POIs with similar signatures can be classified automatically. We demonstrate the possibilities for interactive visual semantic analysis by example of GSM, GPS, and Twitter data. GPS data allow inferring richer semantic information, but temporal signatures alone may be insufficient for interpreting short stops. Twitter data are similar to GSM data but additionally contain message texts, which can help in place interpretation. We plan to develop an intelligent system that learns how to classify personal places and trips while a human analyst visually analyzes and semantically annotates selected subsets of movement data.