Continuous Geo-Social Group Monitoring in Dynamic LBSNs

Geo-social group queries, which return a social cohesive user group with a spatial constraint, have receive significant research interests due to their promising applications for group-based activity planning and scheduling in location-based social networks (LBSNs). However, existing studies on geo-social group queries mostly assume the users are stationary whereas in realistic LBSN application scenarios all users may continuously move over time. Thus, in this paper, we investigate the problem of continuous geo-social groups monitoring (CGSGM) over moving users. A challenge in answering CGSGM queries over moving users is how to efficiently update geo-social groups when users are continuously moving. To address the CGSGM problem, we first propose a baseline algorithm, namely Baseline-BB, which recomputes the new geo-social groups from scratch at each time instance by utilizing a branch and bound (BB) strategy. To improve the inefficiency of BB, we explore a new strategy, called common neighbor or neighbor expanding (CNNE), which expands the common neighbors of edges or the neighbors of users in intermediate groups to quickly produce the valid group combinations. Accordingly, another baseline algorithm, namely Baseline-CNNE, is proposed. As these baseline algorithms do not maintain intermediate results to facilitate further query processing, we develop an incremental algorithm, called incremental monitoring algorithm (IMA), which maintains the support, common neighbors and the neighbors of current users when exploring possible user groups for further updates and query processing. Since IMA requires many times of truss decomposition when processing mutiple-users updates, we propose an improved incremental algorithm, called improved incremental monitoring algorithm (IIMA), which performs truss decompostion only once. Moreover, we design algorithms for handling the social changes that result in insertion/deletion of some edges in the social network. Owing to the challenge in setting, an appropriate monitoring distance, we further study the top $N$N CGSGM problem, which finds top $N$N result groups at each time instance. Finally, we conduct extensive experiments using four real datasets to validate our ideas and evaluate the proposed algorithms.