During the excellent Cultured Data Symposium 2020, Tobin Chodos said something like "since there is no mathematically coherent measure of a "success" in music recommendation, since human love of music is so strange and capricious, you could probably reverse the logic of Spotify's recommender engine and get similarly satisfying results, perhaps more satisfying"
Make a bad Spotify recommender. Like, the worst. Bad vibes anti-recommendations.
This is currently very much a proof of concept. It grabs your top 50 songs (long-term), and then does a "farthest neighbor recommendation" based on the audio features Spotify provides. I restricted myself to the 2019 global most streamed tracks, so I couldn't pick total shit. In other words, it is a recommender system that tries to find music that is popular, but you won't like.
Though, to be honest, that *NYSYNC Christmas song is pretty rough.
You can play with it at http://badplaylist.com
"The point is this. Even if there were some objective criteria that make one artwork better than another, as long as context plays a role in our aesthetic appreciation of art, it's not possible to create a tangible measure for aesthetic quality that works for all places in all times. Whatever statistical techniques, or artificial intelligence tricks, or machine-learning algorithms you deploy, trying to use numbers to latch on to the essence of artistic excellence is like clutching at smoke with your hands."
