RumorSphere: A Framework for Million-scale Agent-based Dynamic Simulation of Rumor Propagation

TL;DR

RumorSphere tackles the challenge of simulating rumor propagation at million-scale by combining LLM-driven core agents with efficient ABMs under a Dynamic Interaction Strategy (DIS) and a Hierarchical Resonance Network (HRN). The approach adaptively activates core agents at conflict boundaries using information cocoon theory and a confusion-based metric, while modeling the rest with Confusion-Adaptive Herding to maintain scalability. Empirical results on real Twitter datasets show substantial reductions in simulation bias (average ~26.5%) and strong temporal alignment (Corr ≈ 0.741), with ablations confirming the essential roles of AO, CAH, and HRN. The framework enables realistic explosive spread and supports counterfactual analyses for mitigation strategies, offering a practical tool for studying misinformation dynamics at scale.

Abstract

Rumor propagation modeling is critical for understanding and mitigating misinformation. Existing approaches combining rule-based regular agents with LLM-driven core agents provide a promising paradigm for large-scale rumor simulation. However, overlooking the dynamic nature of core agents and the importance of network topology on rumor spread significantly undermines the simulation performance. To address these issues, we present RumorSphere, a dynamic and hierarchical resonance framework for effective rumor simulation at the million-agent scale. Considering the dynamic role of core agents in rumor evolution, we propose a multi-agent dynamic interaction strategy based on the information cocoon theory, which adaptively identifies and activates critical core agents at conflict boundaries using LLMs, effectively supporting simulations with millions of agents. In addition, we design a hierarchical resonance network that integrates opinion leaders and localized community structures, enabling more realistic modeling of explosive rumor spread in real-world scenarios. Experiments on real-world datasets show that RumorSphere outperforms state-of-the-art methods, reducing simulation bias by an average of 26.5%.

Figures (7)

• Figure 1: (a) Traditional models are confined by fixed rules and static agent architecture, limiting the effective scalability of the simulation. (b) The proposed RumorSphere adaptively partitions ABM and LLM agents by dynamic interaction strategy for million-scale rumor propagation simulation, significantly reducing the simulation bias.
• Figure 2: The overview of the RumorSphere framework. The social media users are simulated as core and regular agents for rumor propagation simulation. In the simulation process: (1) Dynamic Interaction Strategy adaptively identifies core agents at conflict boundaries and determines agents' interaction patterns dynamically for driving rumor propagation; and (2) Hierarchical Resonance Network facilitates realistic explosive propagation by integrating opinion leaders with tightly-knit local community structures.
• Figure 3: Visual comparison of temporal opinion evolution across three real-world rumor events. The plots illustrate the trajectories generated by RumorSphere and state-of-the-art baselines against the ground truth.
• Figure 4: Token consumption across different scenarios.
• Figure 5: Impact of network structure on opinion dynamics. Colors represent belief intensity from blue at -1 for disbelief to red at +1 for acceptance.