Interpretable, ProbabilisticDivisive Clustering of Large Node-Attributed Networks

Abstract: Clustering of social networks, known as community detection is a fundamental partof social network analysis. A community (also known as a module or cluster) is a setof nodes grouped together according to some characteristic. Traditionally, acommunity has been thought of as a set of nodes that are more densely connectedwith each other than the rest of the network. Introducing node attributes to a socialnetwork, allows for two possible sources of information when clustering thenetwork: The network structure, and the attributes describing the nodes.Traditional community detection methods supporting both these sources ofinformation tend to be computationally complex and the resulting clusters aredifficult to interpret in the sense of what characteristic they were grouped on. Wepresent two methods (probabilistic divisive clustering and top-sampled communitysearch) built on-top of already existing community detection methods (CESNA andFocusCO). Both of our methods aim to detect communities with a specifiedattribute association, yielding interpretable results in a feasible amount of time. Thecommunity detection algorithms our methods are built upon, are applied to differentdatasets in order to examine the runtime performance. We also display how ourproposed methods can be used to detect communities formed around topics ofinterest and how a network can iteratively be clustered in order to detect sub-communities with a specified attribute association. In conjunction with researchabout psychological profiling, we believe that our proposed methods could be usedto detect communities of people having similar psychological profile in online socialnetworks.