ONDOCS: Ordering Nodes to Detect Overlap** Community Structure

Abstract

Finding communities is an important task for the discovery of underlying structures in social networks. While existing approaches give interesting results, they typically neglect the fact that communities may overlap, with some hub nodes participating in multiple communities. Similarly, most methods cannot deal with outliers, which are nodes that belong to no germane communities. The definition of community is still vague and the criterion to locate hubs or outliers varies. Existing approaches usually require guidance in this regard, specified as input parameters, e.g., the number of communities in the network, without much intuition. Here we present a general community definition and a list of requirements for a community mining metric. We review advantages and disadvantages of existing metrics and propose our new metric to quantify the relation between nodes in a social network. We then use the new metric to build a visual data mining system, which first helps the user to achieve appropriate parameter selection by observing initial data visualizations, then detects overlap** community structure from the network while also excluding outliers. Experimental results verify the scalability and accuracy of our approach on real data networks and show its advantages over existing methods that also consider overlaps. An empirical evaluation of our metric demonstrates superior performance over previous measures.