Systematic Investigation of Recent Pre-trained Language Model for Hate Speech Detection in Arabic Tweets

• Published in: ACM transactions on Asian and low-resource language information processing 2024-06
• Main Authors: Daouadi, Kheir Eddine, Boualleg, Yaakoub, Guehairia, Oussama
• Format: Article
• Language: English
• Subjects: Do Not Use This Code, Generate the Correct Terms for Your Paper
• ISSN: 2375-4699; 2375-4702
• DOI: 10.1145/3674970

Today, hate speech classification from Arabic tweets has gained significant interest among global researchers. Different techniques and systems are harnessed to overcome this classification task. However, two main challenges are confronted, the use of handcrafted features and the fact that their performance rate is still limited. We address the hate speech identification from Arabic tweets while providing a deeper comprehension of the capability of a new technique based on transfer learning. Specifically, the accuracy result of traditional machine learning (ML) models is compared with Pre-trained Language Models (PLMs) as well as Deep Learning (DL) models. Experiments on a benchmark dataset show that (1) the multidialectal PLMs outperform monolingual and multilingual ones; (2) the fine-tuning of recent PLMs enhances the performance results of hate speech classification from Arabic tweets. The major contribution of this work lies in achieving promising accuracy results in the Arabic hate speech classification task.