Detection and localization of change points in temporal networks with the aid of stochastic block models

TL;DR

This paper extends a framework for detecting change points in temporal networks by incorporating stochastic block models, comparing multiple methods, and finding that degree-corrected SBMs perform well in sparse networks.

Contribution

It introduces the use of stochastic block models within the change point detection framework and evaluates their effectiveness against other methods.

Findings

No method consistently outperforms others in empirical networks.

Degree-corrected SBM shows better recall, especially in sparse networks.

Methods have varying strengths depending on network sparsity and window size.

Abstract

A framework based on generalized hierarchical random graphs (GHRGs) for the detection of change points in the structure of temporal networks has recently been developed by Peel and Clauset [1]. We build on this methodology and extend it to also include the versatile stochastic block models (SBMs) as a parametric family for reconstructing the empirical networks. We use five different techniques for change point detection on prototypical temporal networks, including empirical and synthetic ones. We find that none of the considered methods can consistently outperform the others when it comes to detecting and locating the expected change points in empirical temporal networks. With respect to the precision and the recall of the results of the change points, we find that the method based on a degree-corrected SBM has better recall properties than other dedicated methods, especially for sparse networks and smaller sliding time window widths.