Identification of Community Structures in Networks Employing a Modified Divisive Algorithm

TL;DR

The paper addresses identifying community structure in networks by leveraging modularity $Q$ as the objective. It introduces a two-stage divisive framework, CCR and CCR-EBR, that combines edge cluster coefficient and edge betweenness with a MoveQ refinement to optimize $Q$. The approach yields competitive or superior modularity on multiple networks (e.g., Zachary's Karate Club, Les Misérables) and improves efficiency by integrating refinement during division. This work offers scalable, accurate community detection in large graphs and outlines avenues for further speedups such as parallel processing and faster edge-measure strategies.

Abstract

In numerous networks, it is vital to identify communities consisting of closely joined groups of individuals. Such communities often reveal the role of the networks or primary properties of the individuals. In this perspective, Newman and Girvan proposed a modularity score (Q) for quantifying the power of community structure and measuring the appropriateness of a division. The Q function has newly become a significant standard. In this paper, the strengths of the Q score and another technique known as the divisive algorithm are combined to enhance the efficiently of the identification of communities from a network. To achieve that goal, we have developed a new algorithm. The simulation results indicated that our algorithm achieved a division with a slightly higher Q score against some conventional methods.

Figures (5)

• Figure 1: A graph of a network (left) and the corresponding dendrogramgenerated by our algorithm (right)
• Figure 2: An example of computing the $MoveQ$ score
• Figure 3: Division of the Zachary’s karate club by our algorithm
• Figure 4: Division of the Les-miserables dataset by the CCR algorithm
• Figure 5: Division of the Les-miserables dataset by the CCR-EBR algorithm