Leveraging Moroccan Arabic Sentiment Analysis Using AraBERT and QARIB

Given the significance of Sentiment Analysis, NLP emerged with studies focusing on this task. The studies being conducted to investigate Arabic Natural Language Processing are not as significant as those in other non-Latin alphabet languages are. This lack of research is due to the rich and complex nature of the Arabic language, as well as the lack of free lexical resources. When it comes to dialectal Arabic, the studies become even more limited, revealing additional challenges. As a result, our paper focuses on Moroccan Arabic Sentiment Analysis by incorporating pre-trained Arabic BERT models -AraBERT and QARIB- throughout the process with various combinations such as SVM, CNN, and fine-tuning the pre-trained model itself. Our proposed system achieved the highest accuracy of 93% while using QARIB model. The research results obtained using our approach are compared with those of other leading methods, showing that our approach is highly effective and can provide valuable guidance for future improvements.