Real-Time Fact-Checking using Retrieval-Augmented Generation and Large Language Models

The rapid spread of misinformation on online plat-forms has motivated extensive research in automated fake news detection. Building on the LingML framework for linguistic-informed fake news detection, we study three transformer-based models—ALBERT, BERTweet, and CT-BERT—for COVID-19 news classification and analyze their performance on a bench-mark dataset. We further extend this line of work by developing a real-time fact-checking system that uses retrieval- augmented generation (RAG) and a large language model (LLM) to verify user-submitted claims based on up-to-date web evidence. Our system retrieves relevant web snippets through a search API, constructs a retrieval-augmented prompt, and queries GPT-4o to produce structured verdicts, confidence scores, and explana-tions, which are stored in a database and exposed through a web interface. Experiments on the CONSTRAINT-style COVID- 19 fake news data and qualitative case studies on real-time news demonstrate the feasibility of combining transformer-based classifiers with RAG-based LLM reasoning for practical fact-checking applications.