Community Cores: Removing Size Bias from Community Detection

Community discovery in social networks has received a significant amount of attention in the social media research community. The techniques developed by the community have become quite adept at identifying the large communities in a network, but often neglect smaller communities. Evaluation techniques also show this bias, as the resolution limit problem in modularity indicates. Small communities, however, account for a higher proportion of a social network’s community membership and reveal important information about the members of these communities. In this work, we introduce a re-weighting method to improve both the overall performance of community detection algorithms and performance on small community detection.