Trajectory-Based Community Detection

Zhongyuan Jiang, Xianyu Chen, Bowen Dong, Junsan Zhang, Jibing Gong, Hui Yan, Zehua Zhang, Jianfeng Ma, Philip S. Yu

Abstract: Community phenomenon is ubiquitous in our social activities. For instance, in a football match, the players are divided into two disjoint teams (i.e., communities) in which the ball is frequently forwarded from one player to another, generating many ball transferring trajectories. It is interesting to do a community detection which is only based on the objective trajectories for some specific purpose such as the fraud player detection. In this brief, we first artificially collect the football trajectories for at least 20 football matches of 2018 FIFA World Cup. Secondly, we build a football transferring network in which the link weight is the number of ball transfers from one player to another. Thirdly, we propose a seed based local bottom-up community detection (LBPCD) method which discovers new team members gradually by maximizing the defined modularity. Finally, we compose experiments on both the collected football data and an email network to demonstrate the effectiveness of the proposed method.