To date, we have used the Community Structure Analysis Framework to study the structure of communities found by multiple algorithms across multiple networks.

  • On the Separability of Structural Classes of Communities. Bruno Abrahao, Sucheta Soundarajan, John Hopcroft, Robert Kleinberg, In Proceedings of the 18th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD'12), Beijing, China, 2012. In this paper, we considered 10 community detection methods as well as annotated communities obtained using network metadata, and demonstrated three important results:

    • First, each class of communities was both remarkably consistent internally and distinct from the other classes
    • Second, the annotated communities most resembled the classes of communities obtained by random-walk-based methods
    • Third, that for most networks, a small set of features could be used to differentiate between the classes.

  • Our paper "A Separability Framework for Analyzing Community Structure" has been accepted with minor revisions to TKDD! Final version coming soon!

  • Also coming soon: an in-depth analysis of the structure of real communities.