Quantifying the Interplay Between Panic Propagation and Misinformation on Social Media Using Large Language Models

Abstract

The digital age has catalyzed a phenomenon where information diffusion occurs at unprecedented velocities, often outpacing the capacity for verification. This paper investigates the symbiotic relationship between panic propagation and the spread of misinformation on social media platforms during crisis events. While traditional sentiment analysis and fact-checking systems have operated in isolation, we propose a novel framework that utilizes Large Language Models (LLMs) to jointly model these phenomena. By leveraging the semantic reasoning capabilities of state-of-the-art LLMs, we quantify the causality and temporal lag between exposure to falsified narratives and the subsequent escalation of collective anxiety. Our methodology introduces the Panic-Misinformation Interaction Index (PMII), a metric derived from high-dimensional embedding spaces, to measure the volatility of public discourse. We evaluate our approach on a massive dataset curated from social media feeds during recent global health emergencies. The results demonstrate that misinformation does not merely accompany panic but acts as a primary accelerant, with a quantifiable amplification factor. Furthermore, our LLM-driven approach outperforms baseline deep learning models in predictive accuracy regarding the trajectory of social hysteria.