Fake News Detection: A Review of Conventional and State-of-the-Art Approaches

Abstract

Digital platforms face an urgent problem because misinformation spreads quickly which threatens to damage information accuracy and public confidence. The research evaluates how traditional machine learning models including Logistic Regression and Support Vector Machine perform against the deep learning method Bidirectional Encoder Representations from Transformers (BERT) for fake news identification tasks. The experimental results demonstrate that machine learning models provide interpretable results with average performance. However, BERT outperforms them by understanding news text semantics and context at a deeper level. The research presents an original approach through the development of contextual feature-generation techniques that improve misinformation classification accuracy. The research demonstrates that transformer models can enhance fake news detection systems at scale while building more effective digital information systems.