Year: 2025
Type: conference-paper
Abstract: This paper presents Arabic hate and hope speech classification using pre-trained language models and advanced data augmentation techniques.We evaluate multiple Arabic BERT variants on 6,889 Arabic text samples labeled as hate speech, hope speech, or not-applicable.Data augmentation through back-translation and LLM-based data generation using few-shot prompting significantly improves performance across all models.We establish strong baselines using ensemble bagging XGBoost alongside traditional machine learning approaches.CAMeLBERT with data augmentation achieves the best Macro-F1 of 0.6868, demonstrating the effectiveness of Arabic-specific models combined with modern augmentation strategies for hate speech detection speech detection.
Authors: Nguyen Minh Bao, Dang Van Thin
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4416037757