Density-based community detection in social networks

Abstract

This paper deals with community detection in social networks using density-based clustering. We compare two well-known concepts for community detection that are implemented as distance functions in the algorithms SCAN [1] and DEN-GRAPH [2], the structural similarity of nodes and the number of interactions between nodes, respectively, in order to evaluate advantages and limitations of these approaches. Additionally, we propose to use a hierarchical approach for clustering in order to get rid of the problem of choosing an appropriate density threshold for community detection, a severe limitation of the applicability and usefulness of the SCAN and DENGRAPH algorithms in real life applications. We conduct all experiments on data sets with different characteristics, particularly Twitter data and Enron data.