A BERT Based Architecture for Detecting Arabic Fake News

The widespread impact of fake news necessitates effective detection methods. However, there’s a research gap concerning the evaluation of Arabic BERT models in this context. This research aims to address this gap by evaluating and comparing the performance of different Arabic BERT models in detecting fake news using the recently published CT23-dataset. As fake news techniques evolve and adapt, it is crucial to remain abreast of these developments to effectively identify and combat misinformation. Therefore, leveraging the CT23-dataset, which encompasses a wide range of Arabic tweets, this study seeks to bridge the research gap and provide valuable insights into the capabilities and limitations of Arabic BERT models for precise fake news detection in the Arabic language. Through rigorous experimentation, the study evaluates the performance of five prominent Arabic BERT models:“Arabic Base BERT," “AraBERTV2.0," “CamelBERT MSA," “ArBERT," and “MarBERT." The outcomes contribute to a broader understanding of fake news detection and underscore the potential of Arabic BERT models in addressing the persistent challenge of identifying and combating fake news in Arabic. Notably, “AraBERTV2.0" emerges as the most accurate model, achieving an impressive 96% accuracy rate. Additionally, the study delves into the disparities among existing BERT models and offers insights to enhance the accuracy of fake news detection in Arabic.