Mining Representative Cohesive Dense Subgraphs

Abstract

Data mining techniques have an important implication in social and biological network analysis, were we're interested in finding related complexes and communities. A modern paradigm for solving this problem involves finding densely connected interacting members such as bound proteins in a PPI. It's important to also consider the properties of members. In the context of social networks, we might be interested in finding groups of friends of similar age and sharing common interests. This information can lead to better targeting for advertising and suggestions. In this paper, we introduce an algorithm that can be applied to mining entity- relationship networks. Our approach discovers relevant subnetworks by considering density among entities as well as their similar attribute properties. We apply two distinct methods of forming subnetworks in order to find as many relevant complexes as possible. In addition, we supply techniques for summarization and reduction of nearly-redundant subnetworks in the results.