Interplay between exogenous triggers and endogenous behavioral changes in contagion processes on social networks

TL;DR

This paper develops a model of opinion dynamics incorporating external triggers and validates it with Twitter data, revealing phase transitions and the impact of educated agents on misinformation spread.

Contribution

It introduces a novel opinion dynamics model with external broadcasting and validates it against real social media data, bridging theory and empirical observation.

Findings

Phase transition in unaware agents' distribution under all-to-all approximation.

Disappearance of phase transition with a minimum fraction of educated agents.

Model reproduces key features of Twitter discussions on vaccine misinformation.

Abstract

In recent years, statistical physics' methodologies have proven extremely successful in offering insights into the mechanisms that govern social interactions. However, the question of whether these models are able to capture trends observed in real-world datasets is hardly addressed in the current literature. With this work we aim at bridging the gap between theoretical modeling and validation with data. In particular, we propose a model for opinion dynamics on a social network in the presence of external triggers, framing the interpretation of the model in the context of misbehavior spreading. We divide our population in aware, unaware and zealot/educated agents. Individuals change their status according to two competing dynamics, referred to as behavioral dynamics and broadcasting. The former accounts for information spreading through contact among individuals whereas broadcasting plays the role of an external agent, modeling the effect of mainstream media outlets. Through both simulations and analytical computations we find that the stationary distribution of the fraction of unaware agents in the system undergoes a phase transition when an all-to-all approximation is considered. Surprisingly, such a phase transition disappears in the presence of a minimum fraction of educated agents. Finally, we validate our model using data collected from the public discussion on Twitter, including millions of posts, about the potential adverse effects of the AstraZeneca vaccine against COVID-19. We show that the intervention of external agents, as accounted for in our model, is able to reproduce some key features that are found in this real-world dataset.