Clustering Spotify Songs Into Genres

This project applies non-negative matrix factorization to Spotify song data for cluster analysis. Content-based recommendation systems rely on a basis of songs that can describe a user’s taste profile; for example, Spotify’s Discover Weekly feature creates a curated playlist of songs that are similar to the genres and preferences of a specific user. This project uses a Spotify “tracks” dataset composed of approximately 600,000 Spotify songs and 19 corresponding audio features, as well as artist name. After conducting non-negative matrix factorization, the songs in the dataset are grouped into 9 distinct clusters, or genres, where each cluster is represented by some specifc audio features.

Code for this project can be found here