Mining trajectory profiles for discovering user communities

Abstract

With the rapid development of positioning techniques (e.g., GPS), users can easily collect their trajectories. Furthermore, with the growing of Web 2.0, some web sites allow users to share their own trajectories. In such web sites, users are able to search trajectories that are interested by users. To provide more insights into these trajectories, in this paper, we target at the problem of discovering communities among users, where users in the same community have similar moving behaviors. Note that moving behaviors are usually represented as trajectory patterns where a user frequently travels. In this paper, we propose a framework to discover communities of users. Explicitly, we adopt a probabilistic suffix tree (abbreviated as PST) as a trajectory profile which truly reflects user moving behavior of a user. In light of trajectory profiles, we further formulate a similarity measurement among trajectory profiles of users. Based on the similarity measurement, we develop algorithm CI (standing for Community Identification) to discover user communities. Furthermore, for the same community, one representative PST is selected. When a new user is added, one could simply derive the similarity measurement by comparing representative PSTs, which is able to efficiently determine which community this new user should join. To evaluate our proposed methods, we conduct experiments on the synthetic dataset generated from one real dataset. Experimental results show that the trajectory profile proposed can effectively reflect user moving behavior, and our proposed methods can accurately identify communities among users.