Eva: Attribute-Aware Network Segmentation

TL;DR

Eva is a low-complexity algorithm that effectively detects attribute-homogeneous communities in networks, balancing attribute similarity with structural modularity across diverse real-world datasets.

Contribution

It introduces a novel bottom-up method for attribute-aware community detection that optimizes both structural and attribute homophily criteria.

Findings

Eva outperforms state-of-the-art methods in attribute homogeneity.

It maintains high modularity while grouping nodes by attributes.

Effective in single and multi-attribute scenarios.

Abstract

Identifying topologically well-defined communities that are also homogeneous w.r.t. attributes carried by the nodes that compose them is a challenging social network analysis task. We address such a problem by introducing Eva, a bottom-up low complexity algorithm designed to identify network hidden mesoscale topologies by optimizing structural and attribute-homophilic clustering criteria. We evaluate the proposed approach on heterogeneous real-world labeled network datasets, such as co-citation, linguistic, and social networks, and compare it with state-of-art community discovery competitors. Experimental results underline that Eva ensures that network nodes are grouped into communities according to their attribute similarity without considerably degrading partition modularity, both in single and multi node-attribute scenarios