From a Tiny Slip to a Giant Leap: An LLM-Based Simulation for Fake News Evolution

ACL ARR 2025 February Submission4930 Authors

Abstract: With the growing spread of misinformation online, understanding how true news evolves into fake news has become crucial for early detection and prevention. A critical yet overlooked issue is that fake news usually originates from distorted facts or intentional creation by malicious actors rather than naturally existing in social networks. Hence, we propose $\textbf{FUSE}$ ($\textbf{F}$ake news evol$\textbf{U}$tion $\textbf{S}$imulation fram$\textbf{E}$work), a novel approach using Large Language Models (LLMs) to simulate this evolution process. We model a social network with four types of LLM agents commonly observed in daily interactions: $\textit{spreaders}$ who propagate information, $\textit{commentators}$ who provide interpretations, $\textit{verifiers}$ who fact-check, and $\textit{bystanders}$ who observe passively. These agents interact under various network structures, engaging in daily belief exchanges and reflections that demonstrate information distortion patterns. To evaluate this previously unexplored area, we develop FUSE-EVAL to measure truth deviation during the evolution process. Results show that FUSE effectively captures fake news evolution patterns and accurately reproduces known fake news, aligning closely with human evaluations. Our findings emphasize that preventing misinformation at its early stages is more effective than intervention after full evolution. We hope our work catalyzes further research on early detection and prevention of fake news:https://anonymous.4open.science/r/FUSE-7022/README.md.

Paper Type: Long

Research Area: Computational Social Science and Cultural Analytics

Research Area Keywords: misinformation detection and analysis, NLP tools for social analysis, quantitative analyses of news and/or social media

Contribution Types: NLP engineering experiment, Data analysis

Languages Studied: English

Submission Number: 4930