Project Introduction

After searching through datasets that are available online, we decided to narrow down our research topic into song genre classification. More specifically, we aim to build a classifier that takes song lyrics as inputs and predicts the genre of the song as outputs.

The baseline model will use Naive Bayes and SVM to study how well we can predict a song genre, based only on lyrics.

For advanced models, we will add other features such as artist, year, song popularity, prevalence of vulgarity etc., to predict genres.

Moreover, we will also take a look at how lyrics changes over time. The main questions that interest us are:

(1) Do genres have distinct lyrical trends? If so, what are these trends and do they allow us predict a song’s genre?

(2) Has the lyrical composition of genres changed over the past several decades? If it has, how has it changed?