A Semi-supervised Approach to Visualizing and Manipulating Overlapping Communities

Abstract

When evaluating a network topology, occasionally data structures cannot be segmented into absolute, heterogeneous groups. There may be a spectrum to the dataset that does not allow for this hard clustering approach and may need to segment using fuzzy/overlapping communities or cliques. Even to this degree, when group members can belong to multiple cliques, there leaves an ever present layer of doubt, noise, and outliers caused by the overlapping clustering algorithms. These imperfections can either be corrected by an expert user to enhance the clustering algorithm or to preserve their own mental models of the communities. Presented is a visualization that models overlapping community membership and provides an interactive interface to facilitate a quick and efficient means of both sorting through large network topologies and preserving the user's mental model of the structure.