Dynamic Clustering in Social Networks using Louvain and Infomap Method

TL;DR

This paper introduces a modified Louvain method for dynamic social network clustering that efficiently updates communities with minimal recomputation, maintaining quality while significantly reducing runtime.

Contribution

It presents a novel incremental approach to community detection that adapts Louvain and Infomap methods for dynamic graphs, avoiding full reruns.

Findings

Runtime decreases significantly with minimal quality loss

Effective handling of graph changes without full recomputation

Validation with a custom graph generator

Abstract

Finding communities or clusters in social networks is a fa- mous topic in social network analysis. Most algorithms are limited to static snapshots, so they cannot handle dynamics within the underlying graph. In this paper we present a modification of the Louvain community detec- tion method to handle changes in the graph without rerunning the full algorithm. Also, we adapted the Louvain greedy approach to optimize the Infomap measure. The main idea is, to recalculate only a small area around the changes. Depending on the graph size and the amount of changes, this yields a massive runtime decrease. As validation data, we provide a graph generator, which produces spe- cific community structures, at given times and also intermediate steps to transform the graph from one to another specific graph. Experiments show that runtime decrease is possible without much loss of quality. These values depend on the reprocessed area inside the graph.