Arabic Hate Speech Identification and Masking in Social Media using Deep Learning Models and Pre-trained Models Fine-tuning

2025-07-31

AI Summary

This study addresses hate speech detection and mitigation in Arabic social media. To this end, we propose an integrated detection-and-sanitization framework: first, a fine-tuned pre-trained Transformer model achieves high-accuracy detection (95% accuracy, 92% Macro F1); second, we formulate hate-term sanitization as a sequence-to-sequence machine translation task—specifically, mapping hateful tokens to length-preserving asterisk-based masks in an end-to-end manner. To our knowledge, this is the first work to formalize Arabic hate speech sanitization as a translation problem, thereby eliminating reliance on hand-crafted rules or lexicon-based constraints. We jointly evaluate mask quality using F1 score and 1-gram BLEU (0.3), demonstrating that our generated masks match the fidelity of state-of-the-art translation systems. The approach significantly enhances the efficacy and practicality of joint detection–sanitization pipelines for Arabic online content moderation.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Object Detection & Categorization

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Web data integration and cleaning
• Responsible Web: Mitigating online harms, including harassment, spam, and fake reviews

Abstract

Hate speech identification in social media has become an increasingly important issue in recent years. In this research, we address two problems: 1) to detect hate speech in Arabic text, 2) to clean a given text from hate speech. The meaning of cleaning here is replacing each bad word with stars based on the number of letters for each word. Regarding the first problem, we conduct several experiments using deep learning models and transformers to determine the best model in terms of the F1 score. Regarding second problem, we consider it as a machine translation task, where the input is a sentence containing dirty text and the output is the same sentence with masking the dirty text. The presented methods achieve the best model in hate speech detection with a 92% Macro F1 score and 95% accuracy. Regarding the text cleaning experiment, the best result in the hate speech masking model reached 0.3 in BLEU score with 1-gram, which is a good result compared with the state of the art machine translation systems.

Problem

• Detect hate speech in Arabic social media text
• Clean text by masking hate speech with stars
• Evaluate deep learning models for optimal performance

Innovation

• Deep learning models for Arabic hate speech detection
• Fine-tuning pre-trained models for text masking
• Machine translation approach for hate speech cleaning