Community Detection for Multilayer Heterogeneous Network

TL;DR

This paper introduces a new spectral clustering method based on a modified Degree-Corrected Stochastic Model for detecting communities in multilayer heterogeneous networks, demonstrated on simulated and real data.

Contribution

It proposes a novel spectral clustering approach for multilayer heterogeneous networks and introduces the BiScore algorithm for bipartite network clustering under DCBM.

Findings

Effective community detection on simulated multilayer data

Successful application to Authorship/Citation network data

BiScore guarantees consistent clustering results under mild conditions

Abstract

Many real world networks consist of multiple types of nodes with edges that are heterogeneous in nature. However, most of the existing work for community detection only focused on homogeneous network consisting of a single layer. In this paper, we propose a modified Degree-Corrected Stochastic Model (DCBM) for modeling multilayer heterogeneous network. We develop a spectral clustering method that can unify the information contained in each sub-network, and demonstrate its efficiency to detect communities on simulated data and on Authorship/Citation network data. As a by-product, we present a novel algorithm called BiScore for clustering bipartite network under DCBM, and show that under mild conditions BiScore is guaranteed to yield consistent results.