A Computational Approach to Author Identification from Bengali Song Lyrics

Abstract

Music is one of the truest forms of art. People listen to music both as a form of entertainment and means of relaxation. Every country or region in the world has its own form and style of music. Bangladesh is no exception as it has a great history of music with a great tradition of song writings over centuries. Although songs are very popular among the enthusiasts, authors of them get little recognition. As a result, author identification from songs, more specifically from lyrics, is an important and realistic possibility. Authorship attribution is one of the ways of identifying the author from a linguistic corpus. This paper demonstrates a guideline to identify the author of a Bengali song from the lyrics of that song using machine learning. It presents the first work on machine learning-based computational approach for author attribution from the lyrics of Bengali songs. Six methods of machine learning were used for the author identification, and high accuracy had been achieved from these methods while applied to the data sets D2A, D4A, and D7A, which were built from Bengali song lyrics. It is observed that the Naive Bayes (NB) classifier provides higher accuracy in comparison with the other methods as it shows 93.9, 85, and 86.7% of accuracy while considering the stop words for our three data sets, respectively.