Abstract

Complex networks play a key role in the evolution of communities and individual decisions community members make. These networks are becoming increasingly heterogeneous, linking many different types of entities. Network analysis and community detection algorithms, however, usually reduce complex networks to homogeneous networks composed of entities of a single type. In the process, they conflate relations between different entity types and loose important structural information. In this paper we describe a generalization of the modularity-based community detection algorithm and apply it to complex, heterogeneous networks. First, we redefine network connectivity in terms of influence, measured by the number of paths of any length that exist between two nodes. We define influence-based modularity and use it to partition a network into communities. We also use influence to measure the relative importance of nodes within the network. Our second contribution is mathematical formalism that allows us to represent complex networks by combining multiple heterogeneous types of evidence within a single model. We apply our approach to standard datasets used in literature and show that exploiting additional sources of evidence corresponding to links between, as well as among, different entity types leads to a better understanding of network structure. Besides identifying network structure, our approach can also identify the most influential members of communities, as well as the “weak ties,” who bridge different communities.

Date

December 3, 2025

Authors

Rumi Ghosh, Kristina Lerman