Autocorrelation properties of temporal networks governed by dynamic node variables

TL;DR

This paper investigates how the evolution of node variables influences the autocorrelation and memory properties of synthetic temporal networks, providing analytical tools for their quantification.

Contribution

It introduces a tractable framework for modeling and analyzing autocorrelation in temporal networks governed by stochastic node-variable dynamics.

Findings

Autocorrelation patterns include power-law and exponential decay.

Methods are applicable to a wide range of temporal network data.

Analytical and simulation approaches are both tractable and effective.

Abstract

We study synthetic temporal networks whose evolution is determined by stochastically evolving node variables - synthetic analogues of, e.g., temporal proximity networks of mobile agents. We quantify the long-timescale correlations of these evolving networks by an autocorrelative measure of edge persistence. Several distinct patterns of autocorrelation arise, including power-law decay and exponential decay, depending on the choice of node-variable dynamics and connection probability function. Our methods are also applicable in wider contexts; our temporal network models are tractable mathematically and in simulation, and our long-term memory quantification is analytically tractable and straightforwardly computable from temporal network data.