Emotion Detection from Bangla Text Using Seven Emotion Classes

Abstract

The focus of this thesis is the categorization of seven emotional states: anger, happiness, surprise, fear, sadness, confusion, and disgust. Additionally, a emotion detection study was carried out in Bangla text. This capability of automatically detecting emotions in text has grown in value with the expansion of digital communication. It can be used in social media analysis, consumer feedback, and mental health assistance. However, the complexity in Bangla morphologically and syntactically rich language are often missed by the existing emotion recognition methods that are mostly built over high resource languages like English. The aim of the study is to develop a classification model that will identify emotion from a single Bangla text sample with high accuracy in overcoming these limitations. A dataset was designed and annotated for the study, and then pre-processing techniques like tokenization, normalization, stemming applied specifically for Bangla. Efficiency in handling classification was tested by several machine learning and deep learning models. Model performance for each category of emotion has been presented with the help of key evaluation measures: precision, recall, F1-score. Confusion matrix is also shown in this paper.