Interpretable Probabilistic Divisive Clustering of Large Node-Attributed Networks

Social network analysis is an important set of techniques that are used in many different areas. One such area is intelligence and law enforcement where social network analysis is used to study various kinds of networks. One of the problems with social networks that are extracted from social media is that easily becomes very large and as a consequence difficult to analyze. Therefore, there is a need for techniques that can divide a large network into smaller communities that are more feasible to analyze. Existing community detection algorithms usually only focus on creating communities based on the underlying networks structure and therefore it can be hard to interpret the meaning of communities.In this work, we present two methods for community detection that allows a user to detect communities with an underlying meaning not only based on the relations in the network but also on attributes of the nodes. Our methods use iterative approaches that allow the user to define meaningful properties and are applicable on large social networks with attributed nodes.