From Statistical Learning to Generative Intelligence in Computational Fake News Detection: A Systematic Appraisal of Critical Issues, Methodological Limitations, and Future Trajectories in Data-Driven

Abstract

The widespread distribution of misleading information on the internet is a significant issue that has prompted the academic community to make enormous efforts to create reliable detection methods. People frequently have complete faith in social media, frequently without questioning the veracity and provenance of the information shared on these sites. Therefore, identifying media-rich false news requires a method that utilizes media components and successfully raises detection accuracy. Since cyberspace is always evolving, there is a need for policies that can both effectively stop spreading of fake news with a lot of media coverage and maintain data systems’ reliability. Therefore, this systematic review presents an analysis based on different methods that are employed in detecting fake news. This study includes the publications from 2018 to 2026. The review methodology of this article includes inclusion and exclusion criteria with a PRISMA flow diagram. Machine learning (ML), deep learning (DL), and large language models (LLM) are the methods used to detect false news. Furthermore, it examines various pre-processing methods and different datasets applied in fake news recognition for determining the effective method. Moreover, the efficiency of various ML and DL methods with different datasets was analyzed. Then, challenges, applications and future directions based on fake news recognition were also described in this study. Researchers can do additional studies depending on the performance of each model and dataset using this information.