A new study likens the spread of ideas to contagions spreading through a network.Credit: Drazen Zigic/ iStockphoto/ Getty Images
Infectious diseases, the uptake of new technologies, and the spread of ideas can all be viewed by scientists as contagions spreading through a network. But some are more complex than others. A two-way interaction is enough to be infected by a pathogen, while changing our mind about something could require a group discussion.
A new study presents a way to recognise whether a contagion process belongs to the simple or to the complex type by observing only a few infection events, and with very little information on the network structure.
“In the real world we often observe a single realisation of contagion processes,” explains Giulia Cencetti, a researcher at Fondazione Bruno Kessler in Trento, and first author of the paper. “Yet the available methods to study them are based on statistics that can be estimated only observing several instances of the same process”.
The authors started from real-world data describing the network structure of social interactions in different contexts: a workplace, a school, a scientific conference, a hospital. “They belong to the SocioPatterns collection where data on physical proximity and face-to-face contacts have been gathered using wearable sensors,” explains Alain Barrat, from CNRS and Aix-Marseille University, a co-author of the study.
The researchers simulated spreading processes on these real networks using four contagion models. A simple one based on pairwise interactions, a threshold one, where a node gets infected if the number of infected neighbors is above a predefined value, and two models where spreading depends on group interactions.
They observed that in all models except the threshold one, the nodes with more connections get infected earlier. The researchers then looked at the correlation between the number of groups to which each node belongs to, and the order in which it gets infected. They found that, in models where contagion is driven by group interactions, the nodes belonging to more groups get infected earlier.
For each network, the researchers defined an algorithm combining these two different types of correlation and trained it on the simulated processes. They then tested it on a new set of processes, using it to guess the kind of contagion model and reaching a very high accuracy.
“This could be a starting point to design early-warning system for fake news diffusion on a social network,” says Barrat.
