TunRoBERTa: A Tunisian Robustly Optimized BERT Approach Model for Sentiment Analysis

Abstract

Sentiment Analysis has grown in importance and popularity due to the proliferation of microblogging sites and the increase in posted comments, tweets, and posts, as it allows for the prediction of people's feelings, thoughts, impressions, and opinions. Sentiment analysis is regarded as one of the most active research areas in NLP. As a result, this tool has piqued the interest of marketing and business firms, government organizations, and society as a whole. Based on that, we propose a Tunisian model in this paper. A robustly optimized BERT approach was used to establish sentiment classification from the Tunisian corpus.

Automated sentiment analysis of visually impaired students’ audio feedback in virtual learning environments

“…The work of Sharma & Prasetyo (2022) , however, reveals great improvement compared to other conventional models like that belonging to Singh et al (2021) and Khaiser, Saad & Mason (2023) , who had shown an accuracy of 85.69% and 78.00% consecutively. Our CNN-based approach, in particular, further improves this performance by measuring the remarkable accuracy of 92.00%, which outperforms the combination proposed in TunRoBERTa-CNN by Antit, Mechti & Faiz (2022) (80.60%) (2022) and the state-of-the-art accuracy of 89.70% reported for the novel CNN-RNN-Attention mechanism of Bas.…”