Mining Pattern Similarity for Mobility Prediction in Location-based Social Networks

Abstract

The widespread use of location-based social networks is making such social media one of the major sources of information about people activities and costumes within urban context, allowing to capture and enhance the comprehension of people behaviour, including human mobility regularities. In that sense, the present work describes a novel approach to predict human mobility by using Twitter data. The approach predict the future location of an individual based on her recent mobility history (like individuals typical mobility routines) and on global mobility in the considered geographic area (e.g., mobility routines of all the Twitter users). The prediction approach is based on a novel trajectory pattern similarity measure that allows to identify the more suitable historic patterns to exploit for the prediction of the user next location. If none of the patterns satisfies the similarity threshold, a set of spatio-temporal features characterizing locations and movements among them are combined in a supervised learning approach based on decision trees. The experimental evaluation, performed on a real-world dataset of tweets posted in London, shows the effectiveness and efficiency of the approach in predicting the user's next places, achieving a remarkable accuracy and precision.