Communities of Dense Weighted Networks

Abstract

Complex networks are intrinsically modular. Resolving small modules is particularly difficult when the network is densely connected; wide variation of link-weights invites additional complexities. In this article we present an algorithm to detect community structure in densely connected weighted networks. First, modularity of the network is calculated by erasing the links having weights smaller than a cutoff q. Then one takes all the disjoint components obtained at q=qm, where the modularity is maximum, and modularize the components individually using Newman Girvan’s algorithm for weighted networks. The performance of the proposed algorithm is evaluated on four different types of network. Initially taking microRNA (miRNA) co-target network of Homo sapiens as an example, we show that this algorithm could reveal miRNA modules which are known to be relevant in biological context. We also demonstrate the algorithm for scientific collaboration network, character interaction network of the novel Les Miserables, neural network of C. elegans and email communication network among employees of a company. In all these cases this new algorithm could efficiently detect all the relevant modules, particularly the small ones which are very strongly connected.