Generating dynamic contact graphs with indirect links

TL;DR

This paper introduces a novel dynamic contact graph model that incorporates indirect links via shared locations at different times, improving the accuracy of network property and diffusion process simulations compared to existing models.

Contribution

The paper presents a new model for generating dynamic contact graphs with indirect links, based on activity-driven networks and empirical data fitting, addressing a key challenge in diffusion modeling.

Findings

Model accurately captures empirical network properties.

Inclusion of indirect links improves diffusion simulation accuracy.

Outperforms existing models that only include direct links.

Abstract

Graph models are widely used to study diffusion processes in contact networks. Recent data-driven research has highlighted the significance of indirect links, where interactions are possible when two nodes visit the same place at different times (SPDT), in determining network structure and diffusion dynamics. However, how to generate dynamic graphs with indirect links for modeling diffusion remains an unsolved challenge. Here, we present a dynamic contact graph model for generating contact networks with direct and indirect links. Our model introduces the concept of multiple concurrently active copies of a node for capturing indirect transmission links. The SPDT graph model builds on activity driven time-varying network modelling for generating dynamic contact networks using simple statistical distributions. This model is fitted with a large city-scale empirical dataset using maximum likelihood estimation methods. Finally, the performance of the model is evaluated by analysing the capability of capturing the network properties observed in empirical graphs constructed using the location updates of a social networking app and simulating SPDT diffusion processes. Our results show that, in comparison to current graph models that only include direct links, our graph model with indirect links match empirical network properties and diffusion dynamics much more closely.