Does anyone know if the spotify recommendation algorithm analyzes the actual music in order to group songs? Or does it only do clustering based on users' listening habits? It seems like incorporating the former could help with discovering new and potentially less well known music, which is an area I'd like to see spotify improve on.
That's too bad, I'd love to have the ability to explicitly trade off the recommendation criteria between music features and user preferences. I understand there's no real incentive for spotify to do this though.
