Verse-Based Emotion Analysis of Bengali Music from Lyrics Using Machine Learning and Neural Network Classifiers

Abstract

Throughout the decades, music has been involved with human emotions and inner imaginations. Along with the overgrowing demands for the overhead reduction of human-aided processes, the urge for automation in the linguistic domain has become a preponderance. Research related to Natural Language Processing (NLP) has been flourished well enough in the case of English and other contemporary modern languages. Although Bengali is an enriched multipurpose language with so wide varieties and precious dialects, very few researches have been conducted on emotion recognition in Bengali music. Besides, the Bengali music archive is blessed with pioneering works of some world-class writers, and in addition to that, modern music enthusiasts’ versatile music predilection has made the task of automation in sentiment analysis quite interesting and challenging. In this study, we introduce a verse-based emotion analysis system for Bengali songs that is able to identify certain emotions from textual data. To appraise the relevant results of our resorted approach, we used several Machine Learning (ML) and Neural Network (NN) models. Eventually, we were able to achieve the best accuracy of 80% and 65% using the Bidirectional Encoder Representations from Transformers (BERT) model for both three and two emotion classes.