New model for emotion detecting from French text using bidirectional long short-term memory

Abstract

Due to the fast growth of social networks, humans have transformed from being general users to creators of network information’s by providing reviews, evaluations, and thoughts on social websites expressing their feelings on various topics. Recently, feedback analysis has become important not only for business owners to improve their products based on user feedback, but also for users to help them select the most suitable products by benefiting from other's experiences. Extracting and identifying human emotional states such as happiness, anger, and worry in texts are targets of emotion analysis due to their importance in providing suggestions for companies and users according to their needs. Although, there has been a lot of work on emotion detection in English text, there is currently lack of research on French text that is because of not existing of French emotion dataset. This paper presents an emotion detection model that integrates the Camembert tokenizer with bidirectional long short-term memory (Bi-LSTM) for emotion detection in French text. The proposed model is trained and validated using a dataset that has been annotated for emotions in French. The proposed model achieved accuracy and an F1-score of 98.66% and 98.66%, respectively, outperforming previous work by 26.36%.