Similarity analysis and modeling in mobile societies

Gautam S. Thakur, Ahmed Helmy, Wei-Jen Hsu
2010 Proceedings of the 5th ACM workshop on Challenged networks - CHANTS '10  
A new generation of "behavior-aware" delay tolerant networks is emerging in what may define future mobile social networks. With the introduction of novel behavior-aware protocols, services and architectures, there is a pressing need to understand and realistically model mobile users behavioral characteristics, their similarity and clustering. Such models are essential for the analysis, performance evaluation, and simulation of future DTNs. This paper addresses issues related to mobile user
more » ... arity, its definition, analysis and modeling. To define similarity, we adopt a behavioral-profile based on users location preferences using their on-line association matrix and its SVD, then calculate the behavioral distance to capture user similarity. This measures the difference of the major spatio-temporal behavioral trends and can be used to cluster users into similarity groups or communities. We then analyze and contrast similarity distributions of mobile user populations in two settings: (i) based on real measurements from four major campuses with over ten thousand users for a month, and (ii) based on existing mobility models, including random direction and time-varying community models. Our results show a rich set of similar communities in real mobile societies with distinct behavioral clusters of users. This is true for all the traces studied, with the trend being consistent over time. Surprisingly, however, we find that the existing mobility models do not explicitly capture similarity and result in homogeneous users that are all similar to each other. Thus the richness and diversity of user behavioral patterns is not captured to any degree in the existing models. These findings strongly suggest that similarity should be explicitly captured in future mobility models, which motivates the need to re-visit mobility modeling to incorporate accurate behavioral models in the future.
doi:10.1145/1859934.1859938 dblp:conf/mobicom/ThakurHH10 fatcat:kwmjfk2mr5h2bnfa3czxejl474