Detecting Overlapping Temporal Community Structure in Time-Evolving Networks

TL;DR

This paper introduces a new method for detecting overlapping community structures in dynamic networks by optimizing a quality function with temporal smoothness, revealing insights into evolving complex systems.

Contribution

It proposes a novel quality function, smoothness constraint, and convex relaxation for identifying overlapping communities over time, with theoretical guarantees and experimental validation.

Findings

Overlapping communities can be detected more accurately with temporal correlation.

The method reveals stable community structures in evolving networks.

Experimental results validate the approach on real and synthetic data.

Abstract

We present a principled approach for detecting overlapping temporal community structure in dynamic networks. Our method is based on the following framework: find the overlapping temporal community structure that maximizes a quality function associated with each snapshot of the network subject to a temporal smoothness constraint. A novel quality function and a smoothness constraint are proposed to handle overlaps, and a new convex relaxation is used to solve the resulting combinatorial optimization problem. We provide theoretical guarantees as well as experimental results that reveal community structure in real and synthetic networks. Our main insight is that certain structures can be identified only when temporal correlation is considered and when communities are allowed to overlap. In general, discovering such overlapping temporal community structure can enhance our understanding of real-world complex networks by revealing the underlying stability behind their seemingly chaotic evolution.