Challenges and Breakthroughs of LLMs in Detecting False Information on Social Media: From DELL to Human-Machine Collaboration

Abstract

Due to the rise of large language models (LLMs) in today's society, the spread of false information has become even more rampant, and the detection and intervention of false information are quite challenging. This paper proposes to use LLMs to detect and intervene in false information on social media. More specifically, the DELL framework based on LLMs provides a systematic multi - stage governance solution. This framework maintains strong generalization performance and offers a clear - structured, operable, and interpretable technical path for applying large language models to false information governance. At the same time, it copes with complex scenarios by integrating multi - modal detection. We also explore the role of graph neural networks in rumor detection and combine graph neural networks with LLMs to form a joint framework for collaborative detection and intervention. This paper provides a comprehensive review of the detection and intervention of false information by LLMs, which can help other researchers better understand the current mainstream methods for detecting and intervening in false information.