Mining movement data to extract personal points of interest: A feature based approach

Due to the widespread of mobile devices in recent years, records of the locations visited by users are common and growing, and the availability of such large amounts of spatio-temporal data opens new challenges to automatically discover valuable knowledge. One aspect that is being studied is the identification of important locations, i.e. places where people spend a fair amount of time during their daily activities; we address it with a novel approach. Our proposed method is organised in two phases: first, a set of candidate stay points is identified by exploiting some state-of-the-art algorithms to filter the GPS-logs; then, the candidate stay points are mapped onto a feature space having as dimensions the area underlying the stay point, its intensity (e.g. the time spent in a location) and its frequency (e.g. the number of total visits). We conjecture that the feature space allows to model aspects/measures that are more semantically related to users and better suited to reason about their similarities and differences than simpler physical measures (e.g. latitude, longitude, and timestamp). An experimental evaluation on the GeoLife public dataset confirms the effectiveness of our approach and sheds some light on the peculiar features and critical issues of location based systems.