Generating surrogate temporal networks from mesoscale building blocks

TL;DR

This paper introduces a method to generate surrogate temporal networks that replicate the complex structural and temporal features of real datasets, enabling better analysis of dynamical processes on networks with limited data.

Contribution

The authors propose a novel approach to create surrogate temporal networks by decomposing original data into local subnetworks and assembling them while preserving large-scale correlations.

Findings

Surrogate networks retain key structural and temporal features of original data.

Dynamical processes on surrogate networks produce similar outcomes to those on real data.

Method successfully applied to social interaction datasets.

Abstract

Surrogate networks can constitute suitable replacements for real networks, in particular to study dynamical processes on networks, when only incomplete or limited datasets are available. As empirical datasets most often present complex features and interplays between structure and temporal evolution, creating surrogate data is however a challenging task, in particular for data describing time-resolved interactions between agents. Here we propose a method to generate surrogate temporal networks that mimic such observed datasets. The method is based on a decomposition of the original dataset into small temporal subnetworks encoding local structures on a short time scale. These are used as building blocks to generate a new synthetic temporal network that will hence inherit the shape of local interactions from the dataset. Moreover, we also take into account larger scale correlations on structural and temporal dimension, using them to inform the process of assembling the building blocks. We showcase the method by generating surrogate networks for several datasets of social interactions and comparing them to the original data on two complementary aspects. First, we show that the surrogate data possess complex structural and temporal features similar to the ones of the original data. Second, we simulate several dynamical processes, describing respectively epidemic spread, opinion formation and emergence of norms in a population, and compare the outcome of these processes on the generated and original datasets. We describe the method in detail and provide an implementation so that it can be easily used in future works based on temporally evolving networks.