[Paper Review] Adversarial Coordination on Social Networks

Chen Hajaj, Sixie Yu+2 more|arXiv (Cornell University)|Aug 3, 2018

Opinion Dynamics and Social Influence42 references3 citations

TL;DR

This paper investigates adversarial coordination on social networks using human subject experiments and data-driven agent-based modeling, showing that communication among network neighbors significantly improves consensus resilience against adversarial nodes, while trusted nodes only help when adversaries are numerous and communication is enabled. Adversarial nodes subtly disrupt coordination by opposing local consensus and sending misleading messages, but remain covert to avoid detection.

ABSTRACT

Decentralized coordination is one of the fundamental challenges for societies and organizations. While extensively explored from a variety of perspectives, one issue which has received limited attention is human coordination in the presence of adversarial agents. We study this problem by situating human subjects as nodes on a network, and endowing each with a role, either regular (with the goal of achieving consensus among all regular players), or adversarial (aiming to prevent consensus among regular players). We show that adversarial nodes are, indeed, quite successful in preventing consensus. However, we demonstrate that having the ability to communicate among network neighbors can considerably improve coordination success, as well as resilience to adversarial nodes. Our analysis of communication suggests that adversarial nodes attempt to exploit this capability for their ends, but do so in a somewhat limited way, perhaps to prevent regular nodes from recognizing their intent. In addition, we show that the presence of trusted nodes generally has limited value, but does help when many adversarial nodes are present and players can communicate. Finally, we use experimental data to develop a computational model of human behavior, and explore a number of additional parametric variations, such as features of network topologies, using the resulting data-driven agent-based model.

Motivation & Objective

• To study decentralized consensus in social networks under adversarial influence, where some agents actively disrupt coordination.
• To evaluate the impact of communication between neighboring nodes on consensus success in the presence of adversarial agents.
• To assess the effectiveness of trusted nodes in improving coordination resilience against adversarial tampering.
• To develop a data-driven agent-based model of human behavior based on experimental results for simulation and parametric analysis.

Proposed method

• Conduct human subject experiments where participants are assigned roles as regular or adversarial agents on networked social structures.
• Implement two key design variables: enabling peer-to-peer communication among neighbors and embedding trusted nodes in the network.
• Collect behavioral data from human subjects to calibrate a computational agent-based model reflecting observed human decision-making patterns.
• Use the calibrated model to simulate consensus outcomes under various network topologies and adversarial configurations.
• Systematically vary network parameters such as density, clustering coefficient, and degree distribution to analyze their impact on consensus rates.
• Optimize adversarial and trusted node placement to evaluate strategic advantages in different network settings.

Experimental results

Research questions

• RQ1 How does enabling communication between neighboring nodes affect consensus success in the presence of adversarial agents?
• RQ2 What is the impact of embedding trusted nodes on coordination resilience when adversarial agents are present?
• RQ3 How do adversarial agents strategically disrupt consensus, and to what extent do they reveal their malicious intent?
• RQ4 How do network topology characteristics—such as density, clustering, and degree distribution—affect consensus rates under adversarial conditions?
• RQ5 To what extent do small parameter changes in agent behavior models or node placement strategies influence consensus outcomes?

Key findings

• Communication among neighbors significantly improves consensus rates, especially as the number of adversarial nodes increases, contrary to prior findings suggesting limited impact.
• Trusted nodes provide limited overall benefit but become valuable when adversarial presence is high and communication is enabled.
• Adversarial nodes actively disrupt coordination by selecting colors opposite to their local neighborhood majority and sending misleading messages, though their actions are deliberately subdued to remain covert.
• Increased network density enhances consensus rates and amplifies the value of trusted nodes, particularly when multiple trusted nodes are present.
• Higher clustering tends to hurt coordination, but this negative effect diminishes as adversarial presence increases, and trusted nodes offer no meaningful improvement in such settings.
• Heavy-tailed degree distributions (e.g., scale-free networks) improve consensus when adversaries are few, but this advantage vanishes with five or more adversaries due to higher likelihood of high-degree nodes being compromised.