An Explainable KG-RAG-Based Approach to Evidence-Based Fake News Detection Using LLMs

Abstract

The advent of the Internet and social media has led to the rapid proliferation of fake news. Current state-of-the-art approaches for evidence-based fake news detection primarily utilize vector-based Retrieval Augmented Generation (RAG) systems. Recent studies have proposed RAG systems that outperform vector-based RAG systems by modeling the document store as a Knowledge Graph (KG). In this work, we investigated the performance of a KG-RAG-based approach for evidence-based fake news detection on the AVeriTeC dataset.