music

How was this built?

I'm constantly on the hunt for new mind-blowing albums, but I couldn't find a recommender where I give it an album I like and get back the kmost similar albums. The song and artist recommenders I found all seemed to use user co-occurrence (how often two songs are played by the same person). That felt like cheating, since it builds similarity out of listeners' habits instead of the music itself.

So in my junior year of college I built recmyrecord, an album recommender that works from the music. Each album is a vector of nfeatures: audio features like valence, danceability, and energy, plus music descriptors from rateyourmusic.com (RYM), weighted by how often users apply them. The albums come from RYM's top 5,000, and RYM is also where I track my listening and find new music (almost too often!).

This map shows the albums from that set I've rated. It uses t-SNE, a dimensionality reduction algorithm, to place albums with similar vectors close together in 2D and push dissimilar ones farther apart.