A density-based framework for community detection in attributed networks

TL;DR

This paper introduces AttDeCoDe, a novel density-based community detection method that integrates node attributes and network structure to identify meaningful communities in attributed networks.

Contribution

AttDeCoDe extends density-based community detection by incorporating attribute-driven density estimation, enabling analysis of communities formed through homophily and leader influence.

Findings

AttDeCoDe effectively captures attribute-driven communities in simulated networks.

The method demonstrates competitive performance on real-world research collaboration data.

AttDeCoDe provides interpretable communities combining structural and attribute information.

Abstract

Community structure in social and collaborative networks often emerges from a complex interplay between structural mechanisms, such as degree heterogeneity and leader-driven attraction, and homophily on node attributes. Existing community detection methods typically focus on these dimensions in isolation, limiting their ability to recover interpretable communities in presence of such mechanisms. In this paper, we propose AttDeCoDe, an attribute-driven extension of a density-based community detection framework, developed to analyse networks where node characteristics play a central role in group formation. Instead of defining density purely from network topology, AttDeCoDe estimates node-wise density in the attribute space, allowing communities to form around attribute-based community representatives while preserving structural connectivity constraints. This approach naturally captures homophily-driven aggregation while remaining sensitive to leader influence. We evaluate the proposed method through a simulation study based on a novel generative model that extends the degree-corrected stochastic block model by incorporating attribute-driven leader attraction, reflecting key features of collaborative research networks. We perform an empirical application to research collaboration data from the Horizon programmes, where organisations are characterised by project-level thematic descriptors. Both results show that AttDeCoDe offers a flexible and interpretable framework for community detection in attributed networks achieving competitive performance relative to topology-based and attribute-assisted benchmarks.