Genre Classification From Lyrical Content
Quick Links
One of the most multi-faceted problems in music is genre classification. This is because genres aren’t something that we can concretely pin down or even necessarily musically define. That’s because musical genres are an expression of our cultural values. What I mean by this is that if you traveled back to the 1920’s and tried to show them music that we, in the 21st century, have defined as “pop”, they would laugh. The categories that we use to separate music into different genres are also often overlapping and interlacing. Take rap for example, within rap there are many different artistic streams - most notable east coast and west coast. As you get deeper and deeper into the sub-genres, you realize that there is virtually no stopping how granular you can get with these definitions. Answering this question is important for many reasons. Most notably, genre classification is the backbone of music recommendation systems. If our system knows that a listener is a big fan of country music, it might be important to understand what country music means so that when we suggest songs, we aren’t suggesting something from the genre of Japanese Industrial Noise (look it up! It’s a real genre!)
For the capstone project of a natural language processing course I had the honor of taking at McGill, our team decided to see if lyrics could be used to classify genre. We experimented with a wide array of machine learning techniques to see if we could tease out differences between the different genres without interacting to the music at all.
