building recommendation systems for classical music has a simple data problem- most recommendation systems (for spotify and others) are based on simple user listening histories that look at "people that listened to X also listened to Y".

this is a problem for classical (and jazz) for two reasons a) these genres are not particularly popular on the platform so there are few unique users and b) the songs are LONG so listening sessions contain fewer songs.

track cooccurance based recs work well for popular genres, but these other genres need a different approach to recs and that's actually where AI could do really well by digging into the unstructured data associated with the tracks (sonic analysis of the song, biographical information about the composer, details about featured soloists, etc) rather than relying on piles of user behavior.