Investigating the Performance of BERT Model for Sentiment Analysis on Moroccan News Comments
Mouaad Errami, Mohamed Amine Ouassil, Rabia Rachidi, Bouchaib Cherradi, Soufiane Hamida, Abdelhadi Raihani
Abstract
Sentiment analysis is the field of determining the attitude or emotions expressed in text, which has become popular due to the rise of social media. AI, ML, and DL have greatly impacted the field of NLP and there have been efforts to improve the accuracy of sentiment analysis using linear models and deep neural networks. The analysis of emotions and feelings in Natural Language Processing is a fascinating area of research that aims to understand people's psychological state through their written opinions. Transformer-based models have been found to be highly effective in sentiment analysis and are currently the best option for many languages. However, sentiment analysis in Arabic still requires improvement, particularly in the tokenization stage. This article conducts a comparative study of various Bidirectional Encoder Representations from Transformers (BERT) models and found that MARBERT is the top performer in terms of classification quality and accuracy compared to other Arabic BERT models.