A song emotion identification system from lyrics using heterogeneous ensemble learning

The number of songs is increasing at an explosive rate which has led to the development of automatic song categorization systems. Songs are categorized in different ways like genre, artist, beats per minute (BPM), etc. It is also very important to categorize songs based on their emotional content as the audience often prefers songs of a particular mood. The lyrics of a song become available online instantly after the release of the song itself. This sets the stage for an automated song classification system based on lyrics. In this paper, a system is presented to address this problem. The system was tested with 400 songs from 2 categories namely happy and sad. The dataset was composed of disparate artists, genres, and timelines wherein the highest accuracy of 83.25% was obtained using a heterogeneous ensemble learning-based approach.