Fact-Augmented Reasoning Model for Fake News Detection

The spread of fake news poses significant societal risks. While existing detection methods often focus on analyzing news content, they generally lack mechanisms to assess consistency with real-world facts explicitly. Factual verification approaches, on the other hand, rely on static, pre-constructed evidence sets and are limited in their ability to proactively retrieve and evaluate facts. To address these limitations, we propose FAR-FD (Fact-Augmented Reasoning Model for Fake News Detection), a novel approach that leverages large language models (LLMs) and the Retrieval-Augmented Generation (RAG) framework. Specifically, we introduce the Factual Information Retrieval and Evaluation processes to proactively acquire and evaluate external factual information to ensure the validity of factual information. Subsequently, we obtain explainable reasoning-based factual text through the LLM reasoning process, which is fed into an Expert Model for final classification. Extensive experimental results on two public benchmark datasets validate the validity and superior performance of our proposed FAR-FD over state-of-the-art detection models.