Arabic Offensive Language Detection Based on Zero-Shot Prompting LLM in Social Media

The swift growth of user-created material on digital platforms has heightened the need for efficient automated methods to detect and categorize offensive language. This article presents a method for categorizing offensive Arabic content on a large scale using LLMs. We explore the challenges of understanding the Arabic language because of its intricate morphology, variations in dialects, and subtle nuances in contexts. We assess zero-shot prompting to distinguish between offensive and non-offensive Arabic language. Additionally, we consider the caliber of features extracted specifically for this task by leveraging the power of pre-trained LLMs. Our findings show that LLM-based models with zero-shot prompting on Arabic-language platforms provide rich resources for feature extraction and contribute to advancing Arabic offensive content detection.