A Large Language Model-based Fake News Detection Framework with RAG Fact-Checking

The widespread dissemination of online misinformation poses significant threats to the public interest, highlighting the urgent need for effective fake news detection. In the era of Large Language Models (LLMs), the rise of AI-generated fake news has intensified this issue, making misinformation more pervasive and harder to control. While fact-checking offers a promising solution by leveraging external knowledge, efficiently linking claims within news articles to relevant external facts remains a significant challenge. To address this, we propose a mis-information detection framework FCRV (Full-Context Retrieval and Verification) that constructs a "full-context" for news articles by integrating LLM-based claim extraction with Retrieval-Augmented Generation (RAG) for fact-checking. We implemented an LLM pipeline for human-like extraction of key claims from datasets, significantly improving extraction quality over traditional methods. Our retrieval workflow effectively detects fictitious entities prevalent in AI-generated news by identifying claims lacking a basis in reality. Experiments across multiple human-generated and AI-generated datasets demonstrate that verifying news using this "full-context" approach leads to more stable and robust fake news detection, enhancing scalability, accuracy, and the model’s ability to handle AI-generated content.