Psychology-driven LLM Agents for Explainable Panic Prediction on Social Media during Sudden Disaster Events

TL;DR

This work introduces PsychoAgent, a psychology-driven generative agent framework for explainable panic prediction on social media during sudden disasters. It pairs a novel COPE dataset, built via human–LLM collaboration, with a four-stage LLM-based agent that models disaster perception, risk perception, panic arousal, and posting behavior under psychological theory, enabling mechanistic interpretability. Empirical results show PsychoAgent improves panic prediction accuracy and AUC by substantial margins over a range of baselines, and ablations confirm the necessity of its risk sensing, arousal, and multi-expert validation components. The framework shifts emphasis from opaque data-fitting to interpretable, role-based simulation of psychological chains, with implications for real-time crisis management and targeted interventions. Future work could extend this approach to modeling panic propagation in networks and developing real-time mitigation strategies.

Abstract

During sudden disaster events, accurately predicting public panic sentiment on social media is crucial for proactive governance and crisis management. Current efforts on this problem face three main challenges: lack of finely annotated data hinders emotion prediction studies, unmodeled risk perception causes prediction inaccuracies, and insufficient interpretability of panic formation mechanisms. We address these issues by proposing a Psychology-driven generative Agent framework (PsychoAgent) for explainable panic prediction based on emotion arousal theory. Specifically, we first construct a fine-grained open panic emotion dataset (namely COPE) via human-large language models (LLMs) collaboration to mitigate semantic bias. Then, we develop a framework integrating cross-domain heterogeneous data grounded in psychological mechanisms to model risk perception and cognitive differences in emotion generation. To enhance interpretability, we design an LLM-based role-playing agent that simulates individual psychological chains through dedicatedly designed prompts. Experimental results on our annotated dataset show that PsychoAgent improves panic emotion prediction performance by 12.6% to 21.7% compared to baseline models. Furthermore, the explainability and generalization of our approach is validated. Crucially, this represents a paradigm shift from opaque "data-driven fitting" to transparent "role-based simulation with mechanistic interpretation" for panic emotion prediction during emergencies. Our implementation is publicly available at: https://anonymous.4open.science/r/PsychoAgent-19DD.

Figures (6)

• Figure 1: Path of panic formation. The physical domain supplies an objective benchmark, while the information domain dynamically adjusts cognitive inputs. The cognitive domain processes these inputs through individual constraints and risk perception, triggering panic emotions that ultimately materialize as observable social media behaviors through the social domain.
• Figure 2: Overview of the proposed panic emotion prediction framework——PsychoAgent.
• Figure 3: Detailed stages for building LLM-based agent driven by CoT in the cognitive domain module.
• Figure 4: Crowdsourced validation protocol for panic emotion annotation.
• Figure 5: Non-panic User Prediction Case.