Enhancing News Credibility Detection in Arabic: A Study of Attentional Bidirectional LSTM

Abstract

Detecting fake news continues to be a major challenge in the digital era, especially in languages with significant linguistic diversity such as Arabic, which is highly ambiguous, morphologically rich, and has numerous dialects. This study presents an enhanced method for detecting Arabic fake news by leveraging an Attentional Bidirectional Long Short-Term Memory (AttBiLSTM) model. Our proposed model integrates an attentional mechanism with BiLSTM to improve classification accuracy. Evaluated on the benchmark AFND Fake News dataset, our approach demonstrates significant improvements in identifying news credibility, achieving an accuracy of $\mathbf{7 9 . 2 0 \%}$. This research advances deep learning methods for fake news detection by offering an effective solution tailored to Arabic language contexts.