Sentiment Analysis of Bengali Music based on various Audio Features: An analysis of Machine Learning and Deep Learning Methods

Sentiment analysis is a method used to determine the emotional tone or mood conveyed in a text or work of literature. Music functions as a constructive medium for emotional expression, providing a powerful means to communicate and convey feelings. Recently, music sentiment analysis has emerged as a popular method for curating and recommending music to listeners based on their emotional state. Despite the abundant literary legacy of the Bengali language, there are only a limited number of notable works that effectively accomplish the desired objective, and the number of sentiment categories is quite low. Furthermore, these efforts rely exclusively on music lyrics, which may not always be an optimal approach. This is because many lines in a song may lack a literal meaning, making it challenging for classifiers to accurately assign them to the appropriate sentiment category. Furthermore, each song possesses inherent audio characteristics. Therefore, in this research, we propose a novel approach aimed to categorize music sentiments into five distinct classes by utilizing these fundamental audio characteristics. Furthermore, we utilized our own dataset to accomplish the desired outcome. We have employed machine learning and deep learning classifiers to accurately categorize the sentiments of Bengali music into appropriate groups. We used suitable metrics to assess the efficiency of our models. In addition, we have conducted an analysis to determine which intrinsic audio characteristics are most significant in relation to the sentiment categories. Furthermore, our models have demonstrated exceptional performance, with a peak accuracy of 76.79%.