DualMind: Towards Understanding Cognitive-Affective Cascades in Public Opinion Dissemination via Multi-Agent Simulation

TL;DR

DualMind tackles the challenge of forecasting public opinion during PR crises by modeling the coupled evolution of cognitive beliefs and transient affect on platform-specific social networks. It introduces a dual-state agent architecture with a slow semantic persona $z_i^t$ and a fast affect state $r_i^t$, updated through a gated coupling and episodic memory, and a PAACM decision policy that integrates content, affect, and memory. The system is instantiated as an LLM-driven multi-agent platform and evaluated on 15 real-world crises after August 2024, achieving high trajectory fidelity ($r \approx 0.78$) and low final-state divergence ($\mathrm{JSD} \approx 0.27$) compared to SOTA baselines, indicating strong cross-cultural validity. The work offers a practical tool for proactive crisis management and strategy testing, with publicly available code to enable replication and extension in real-world PR planning.

Abstract

Forecasting public opinion during PR crises is challenging, as existing frameworks often overlook the interaction between transient affective responses and persistent cognitive beliefs. To address this, we propose DualMind, an LLM-driven multi-agent platform designed to model this dual-component interplay. We evaluate the system on 15 real-world crises occurring post-August 2024 using social media data as ground truth. Empirical results demonstrate that DualMind faithfully reconstructs opinion trajectories, significantly outperforming state-of-the-art baselines. This work offers a high-fidelity tool for proactive crisis management. Code is available at https://github.com/EonHao/DualMind.

Figures (5)

• Figure 1: Overview of the DualMind pipeline.
• Figure 2: User interface of Scenario 1 (Real-world Cases). (A) The user selects a real-world event from the "Select PR Case Study" window to set the ground truth; (B) The "Comparative Analysis Report" provides immediate, top-level fidelity metrics (e.g., overall Pearson's $r$ and JSD); (C) The "Metrics Trend Analysis" graph allows the user to assess the fidelity of the opinion trajectory as it evolves over simulation rounds.
• Figure 3: User interface for Scenario 2. (A) The user iteratively inputs natural language PR strategies in the left panel; (B) The center panel provides a real-time visualization of the opinion propagation network; (C) The right "Message Propagation History" panel tracks agent posts. Users analyze feedback from (C) to refine strategies in (A) and observe the new strategy's diffusion in (B).
• Figure 4: Some major public relations event cases used in assessments from around the world.
• Figure 5: Fidelity comparison. DualMind achieves the highest trajectory similarity and the lowest outcome divergence.