FAD-SA-GRU achieves 93.2% accuracy on Algerian Darija hate speech detection

A new hybrid architecture named FAD-SA-GRU has been developed to improve automatic hate speech detection in the Algerian Arabic dialect (Darija) on social media. The model addresses the linguistic diversity of the dialect by combining semantic representations from DZ FastText, DZ AraVec, and DziriBERT through multi-embedding fusion, followed by a self-attention-enhanced GRU encoder.

• FAD-SA-GRU outperforms traditional machine learning, recurrent neural networks, and Transformer-based baselines like multilingual BERT.
• The model achieved 93.2% accuracy, 93.4% precision, 91.0% recall, 92.1% F1-score, and 97.0% ROC-AUC on an annotated dataset of Algerian Darija social media comments.
• Results demonstrate the effectiveness of combining complementary embedding representations with attention-based sequence modeling for robust detection in low-resource dialectal Arabic.