Natural clustering: the modularity approach

TL;DR

This paper explores how modularity, a concept from network theory, can be generalized to assess clustering quality, especially in algorithms inspired by Statistical Mechanics, providing a new perspective on clustering evaluation.

Contribution

It introduces a generalized modularity measure for clustering, connecting network theory and Statistical Mechanics to improve clustering solution assessment.

Findings

Modularity can be effectively used as a clustering quality indicator.

The approach bridges network theory and physical system analogies in clustering.

Generalized modularity enhances the evaluation of clustering solutions.

Abstract

We show that modularity, a quantity introduced in the study of networked systems, can be generalized and used in the clustering problem as an indicator for the quality of the solution. The introduction of this measure arises very naturally in the case of clustering algorithms that are rooted in Statistical Mechanics and use the analogy with a physical system.