Synthesizing Routes for Low Sampling Trajectories with Absorbing Markov Chains

Abstract

The trajectory research has been an attractive and challenging topic which blooms various interesting location based services. How to synthesize routes by utilizing the previous users' GPS trajectories is a critical problem. Unfortunately, most existing approaches focus on only spatial factors and deal with high sampling GPS data, but low-sampling trajectories are very common in real application scenarios. This paper studies a new solution to synthesize routes between locations by utilizing the knowledge of previous users' low-sampling trajectories to fulfill their spatial queries' needs. We provide a thorough treatment on this problem from complexity to algorithms. (1)We propose a shared-nearest-neighbor (SNN) density based algorithm to retrieve a transfer network, which simplifies the problem and shows all possible movements of users. (2) We introduce three algorithms to synthesize route: an inverted-list baseline algorithm, a turning-edge maximum probability product algorithm and a hub node transferring algorithm using an Absorbing Markov Chain model. (3) By using real-life data, we experimentally verify the effectiveness and the efficiency of our three algorithms.