Fake News Detection System

Abstract

This paper presents a hybrid, explainable, and cross-domain Fake News Detection System that combines deep learning, natural language processing, metadata analysis, and real-time verification. The proposed system uses a fine-tuned RoBERTa transformer for language classification, LIME for interpretability, Google Fact Check Tools API for factual claim verification, and contextual reasoning to analyze potentially misleading content. The system accepts textual and URL-based inputs, integrates multiple evidence sources, evaluates source credibility, and produces interpretable True, False, or Mixed verdicts. The implementation uses Python, PyTorch, Hugging Face Transformers, Flask, MongoDB, and external news and fact-checking APIs.