Agent-Based Models for Assessing Social Influence Strategies

Abstract

Motivated by the increasing attention given to automated information campaigns and their potential to influence information ecosystems online, we argue that agent-based models of opinion dynamics provide a useful environment for understanding and assessing social influence strategies. This approach allows us to build theory about the efficacy of various influence strategies, forces us to be precise and rigorous about our assumptions surrounding such strategies, and highlights potential gaps in existing models. We present a case study illustrating these points in which we adapt a strategy, namely, amplification, commonly employed by so-called ‘bots’ within social media. We treat it as a simple agent strategy situated within three models of opinion dynamics using three different mechanisms of social influence. We present early findings from this work suggesting that a simple amplification strategy is only successful in cases where it is assumed that any given agent is capable of being influenced by almost any other agent, and is likewise unsuccessful in cases that assume agents have more restrictive criteria for who may influence them. The outcomes of this case study suggest ways in which the amplification strategy can be made more robust, and thus more relevant for extrapolating to real-world strategies. We discuss how this methodology might be applied to more sophisticated strategies and the broader benefits of this approach as a complement to empirical methods.