Evolution of Communities with Focus on Stability

Abstract

Community detection is an important tool for analyzing the social graph of mobile phone users. The problem of finding communities in static graphs has been widely studied. However, since mobile social networks evolve over time, static graph algorithms are not sufficient. To be useful in practice (e.g. when used by a telecom analyst), the stability of the partitions becomes critical. We tackle this particular use case in this paper: tracking evolution of communities in dynamic scenarios with focus on stability. We propose two modifications to a widely used static community detection algorithm: we introduce fixed nodes and preferential attachment to pre-existing communities. We then describe experiments to study the stability and quality of the resulting partitions on real-world social networks, represented by monthly call graphs for millions of subscribers.