Past to Present - The Evolution of Fake News Detection Techniques

Abstract

The phenomenon of creation and dissemination of fake news is growing at a pace that often surpasses the development of effective mechanisms for its detection. With rapid advancements in computational technologies, researchers have explored a wide range of approaches to detect and mitigate the spread of misinformation on online social networks. Initial efforts relied on traditional machine learning models that focused on surface-level textual features. The emergence of deep learning and large language models has enabled more sophisticated detection techniques, capable of capturing complex linguistic nuances and patterns, helping distinguish true content from false. This study aims to systematically analyze and compare the performance, strengths, and limitations of three major groups of models, namely Machine Learning, Deep Learning, and Language Models, to assess their robustness and effectiveness in the detection of fake news. Lastly, the execution time of each model is analyzed to assess the computational cost associated with their training and inference, which is a critical factor when evaluating their practicality for real-world deployment.