AI-Driven Explainable Fake News Detection Using Large Language Models, Evidence Retrieval and Source-Aware Reasoning

Abstract

The rapid growth of digital media has made false and misleading information easier to create and distribute. Earlier fake-news detection systems largely relied on machine-learning classifiers, text vectorisation, and ensemble voting. Recent work has introduced BERT, large language models (LLMs) and retrieval-augmented generation (RAG), but a practical detector still needs reliable evidence, source awareness, uncertainty handling and understandable explanations. This paper proposes an AI-driven explainable framework that does not depend only on a trained classifier. The proposed system decomposes a news item into verifiable claims, retrieves supporting and contradicting evidence from multiple sources, evaluates source credibility and information freshness, performs semantic reasoning with an LLM, and combines the evidence into an evidence-weighted confidence score. Instead of forcing every input into a binary decision, the framework can return Real, Fake, or Uncertain together with the key evidence and a human-readable explanation. The paper presents the architecture, methodology, research gap, algorithm, evaluation plan and implementation design. The proposed approach is intended to improve transparency and robustness for both conventional misinformation and AI-assisted fabricated content.