Improved Emotion Detection Framework for Arabic Text using Transformer Models *

Abstract: Emotion detection in text is a challenging task with various applications in natural language processing and psychology. In recent years, there has been increasing interest in developing algorithms for detecting emotions in Arabic text, given the importance of this language and the lack of resources in this domain. This paper proposes the use of transformer-based models for Arabic emotion detection in text. We use the emotone_ar dataset, a resource for the development and evaluation of algorithms and techniques for emotion detection in Arabic. The proposed model is based on transformers for encoding contextual information in text and classify emotions based on this encoded representation. We evaluate the performance of our model on the emotone_ar dataset and compare our results to previous methods for emotion detection. Our model achieves an accuracy of 74.16% and an F1 score of 0.7406 on the test set, outperforming previous methods for emotion detection on this dataset. We also compare our results to the performance of a Naïve Bayes classifier and show that our approach significantly outperforms this baseline. These results demonstrate the effectiveness of transformer-based models for emotion detection in Arabic text and highlight the potential for further improvements in this area.