An Empirical Study of Community Detection Algorithms on Social and Road Networks

TL;DR

This study empirically compares multiple community detection algorithms on social and road networks with different structural properties, providing insights into their performance using diverse metrics.

Contribution

It offers a comprehensive evaluation of community detection algorithms on different network types, highlighting their strengths and limitations.

Findings

Algorithms vary in effectiveness depending on network density and sparsity.

Certain algorithms perform better in social networks than in road networks.

Evaluation metrics reveal nuanced differences in community structures.

Abstract

Community detection in social networks is a problem with considerable interest, since, discovering communities reveals hidden information about networks. There exist many algorithms to detect inherent community structures and recently few of them are investigated on social networks. However, it is non-trivial to decide the best approach in the presence of diverse nature of graphs, in terms of density and sparsity, and inadequate analysis of the results. Therefore, in this study, we analyze and compare various algorithms to detect communities in two networks, namely social and road networks, with varying structural properties. The algorithms under consideration are evaluated with unique metrics for internal and external connectivity of communities that includes internal density, average degree, cut ratio, conductance, normalized cut, and average Jaccard Index. The evaluation results revealed key insights about selected algorithms and underlying community structures.