Arabic Fake News Detection on X(Twitter) Using Bi-LSTM Algorithm and BERT Embedding

The widespread popularity of social media networks facilitates the communication and sharing of events, activities, and content generated by users, which positively impacts our daily lives. However, this also facilitates the propagation of fake news among users. Consequently, the significant influence of fake news on social media can have a domino effect on both individuals and societies that cannot be underestimated, creating an imperative need to automatically detect fake news to limit its spread. Hence, there is a strong desire within the research community to develop efficient systems for fake news detection. The automatic detection of fake news is a very challenging task, and the level of challenge has increased, especially in rich languages such as Arabic, owing to its different formats, dialects, script forms, and semantic structures. However, deep learning techniques have proven their ability to handle natural language processing (NLP) tasks. This study presents a deep learning approach for detecting fake news in Arabic on the X platform (formerly known as Twitter). The approach proposed in this study utilizes the bidirectional long short-term memory (bi-LSTM) algorithm for fake news classification, combined with the AraBERT word embedding model for feature extraction, following an ablation study that evaluated AraBERT, MARBERT, and ARBERT models across three preprocessing variants: original text of tweets without cleaning, lightly cleaned text, and text after applying the stemming technique. Based on the results of the ablation study, the AraBERT model with the original text was selected over other Arabic BERT-based models owing to its superior performance while maintaining efficiency. The experimental results demonstrate the effectiveness of the model, achieving an accuracy of 0.87, an area under the curve (AUC) of 0.96, and an F1-score of 0.88 in distinguishing fake news from real news. To ensure reproducibility and support for future research, the implementation code encompassing data preprocessing, exploration, model training, and evaluation is publicly accessible at https://github.com/Asiri-Hanan/Arabic-Fake-News-Detection-on-X-Twitter-using-bi-LSTM-Algorithm-and-BERT-Embedding-.git.