A Statistical Modelling Approach to Detecting Community in Networks

TL;DR

This paper introduces a new statistical modeling approach for detecting communities in networks by analyzing information spread, offering a robust, scalable method that integrates supplementary data and outperforms heuristic techniques.

Contribution

The paper presents a novel latent community detection technique based on information spread modeling, improving robustness and scalability over existing heuristic methods.

Findings

Effective in real-world networks like Zachary karate club

Handles supplementary information such as communication patterns

Demonstrates robustness and scalability

Abstract

There has been considerable recent interest in algorithms for finding communities in networks - groups of vertex within which connections are dense (frequent), but between which connections are sparser (rare). Most of the current literature advocates an heuristic approach to the removal of the edges (i.e., removing the links that are less significant using a well-designed function). In this article, we will investigate a technique for uncovering latent communities using a new modelling approach, based on how information spread within a network. It will prove to be easy to use, robust and scalable. It makes supplementary information related to the network/community structure (different communications, consecutive observations) easier to integrate. We will demonstrate the efficiency of our approach by providing some illustrating real-world applications, like the famous Zachary karate club, or the Amazon political books buyers network.