Context-based location clustering and prediction using conditional random fields

Roland Assam, Thomas Seidl

Abstract: The embrace of pervasive devices accounts for the production of a massive amount of location data. While multitudes of algorithms have been used for location clustering, most of them focus on the proximity clustering of locations rather than on their location contexts. In this work, we propose a novel context-based location clustering technique that clusters locations with similar context by solely using raw GPS data from multi-user trajectories. We introduce a new similarity measure that infers the location context and utilize the inferred contexts during clustering. In addition, we propose a predictive model that employs Conditional Random Fields (CRF), context-based location clusters and social ties for future location prediction. We show the strength and efficiency of our techniques through numerous experiments on two real datasets. Our empirical evaluations demonstrate that our approach performs better than a state-of-the-art work.