Context based Emotion Recognition from Bengali Text using Transformers

Every individual’s everyday life has become a routine of expressing their ideas, opinions, emotions, and experiences through social networking sites and platforms on the internetin this era of rapid technological innovation. These points of view can be utilized to develop strategies for increasing efficiency in a range of fields, including business, politics, research, and analysis. Natural language processing (NLP) uses emotion detection to automatically monitor, evaluate, and categorize people’s ideas and opinions in order to get a sense of how they feel. To date, a substantial amount of study has been done on the Emotion Recognition of the English language, with notable results. Unfortunately, there has been a scarcity of research on the Bangla language in the subject of Emotion Recognition. Despite the fact that social networks have increased the popularity of romanized Bangla among Bangla speakers, there is even less research on romanized Bangla text. As a result, the focus of this study was on emotion recognition for both raw and romanized Bangla texts. A corpus of romanized Bangla texts was created from a raw Bangla feeling corpus in this study. Datasets of military, medical, religious and general context are collected and tested with the Bidirectional Encoder Representations from Transformers. Finally, the outputs are studied and evaluated.