Moroccan Sentiment Classification Based on DarijaBERT and Stacking Deep Learning Model

Internet access is required for information sharing in both personal and professional settings. Business and Marketing research are all interested in sentiment classification currently. With millions of ratings and comments made every day, the Internet and social media are quickly becoming a great resource and gold mine for businesses and organizations looking to improve their management, production, and marketing. Dialectal Arabic studies are becoming less popular due to their numerous obstacles. Nonetheless, we intend to investigate the sentiment analysis of the Moroccan dialect (MD) dataset using a novel approach based on DarijaBERT embedding and stacked deep learning method using: LSTM, GRU, CNN, and BiLSTM, which provide superior results on two dialect datasets in terms of accuracy, ROC-AUC, and Cohen’s Kappa. Finally, we looked at Modern Standard Arabic (MSA) and Moroccan dialect (MD) Facebook comments from Transport firm in Morocco (CTM) customers to identify customer satisfaction as positive, neutral, or negative comments.