SMOPAT: Mining semantic mobility patterns from trajectories of private vehicles

With the increasing use of private vehicles with positioning services, GPS trajectory data of vehicles has become one of the major sources of big data about urban life. Existing studies on mobility pattern mining from trajectories share a common limitation, i.e., they fail to capture the semantics of trajectories. Automatic derivation of semantic information for every trajectory is a challenging task. In this paper, we propose an approach, called SMOPAT (Semantic MObility PATterns), for mining spatial-temporal semantic mobility patterns from trajectories of private vehicles. We design a probabilistic generative model with latent variables to characterize the semantic mobility of vehicles. Based on the model, SMOPAT labels each location in a trajectory with a visit purpose by using a polynomial-time dynamic programming algorithm. It then employs an efficient algorithm to find the most frequent semantic mobility patterns. We evaluate our approach on a large data set of real trajectories of private vehicles spanning a time duration of over ten months with 114 million records in Shanghai, China. The experimental results show that our approach produces meaningful patterns and outperforms the two competing methods in terms of diversity, coherence, and coverage.