Fake News Detection using NLP: A Hybrid Approach Combining BERT and Graph Neural Networks

Abstract

This rapid spread of false information on digital platforms is creating haul urgent social dilemmas for which corrective automated systems are needed to combat fake news. We propose a hybrid framework that integrates a BERT-based language processing with a contextuality imposed by Graph Neural Networks. It lays a model of the examination over both the grid contents of news articles and metadata associated with credibility of authorship and social networking behaviours, tagging them as genuine or fake. When tested on datasets like LIAR and FakeNewsNet, the model achieves 93.2% accuracy, outperforming traditional methods such as LSTMs and SVMs. Importantly, this framework is robust against deceptive language and adapts well to different datasets. It also integrates explainable AI
techniques, highlighting phrases (like "miracle cure") that signal fake news, making it suitable for real-world applications like social media moderation.