Consistency landscape of network communities

TL;DR

This paper introduces a framework to analyze the stability and reliability of community detection in networks by examining the consistency landscape across different resolutions, addressing the challenge of choosing appropriate community scales.

Contribution

It proposes a novel method leveraging consistency measures to quantify community structure reliability across resolutions, improving upon heuristic approaches.

Findings

Effectively visualizes community structure stability

Identifies optimal community resolutions in synthetic and real networks

Provides insights into fundamental properties of network communities

Abstract

The concept of community detection has long been used as a key device for handling the mesoscale structures in networks. Suitably conducted community detection reveals various embedded informative substructures of network topology. However, regarding the practical usage of community detection, it has always been a tricky problem to assign a reasonable community resolution for networks of interest. Because of the absence of the unanimously accepted criterion, most of the previous studies utilized rather ad hoc heuristics to decide the community resolution. In this work, we harness the concept of consistency in community structures of networks to provide the overall community resolution landscape of networks, which we eventually take to quantify the reliability of detected communities for a given resolution parameter. More precisely, we exploit the ambiguity in the results of stochastic detection algorithms and suggest a method that denotes the relative validity of community structures in regard to their stability of global and local inconsistency measures using multiple detection processes. Applying our framework to synthetic and real networks, we confirm that it effectively displays insightful fundamental aspects of community structures.