3 ms·

I keep hearing from people how good is Spotify "Discovery" playlist. I've been a paid customer for 5+ years and their recommendations have a hit rate of 2%. I l

I keep hearing from people how good is Spotify "Discovery" playlist. I've been a paid customer for 5+ years and their recommendations have a hit rate of 2%. I listen to a lot of different genres, artists, mostly dictated by my mood. Today it was some easy Dire Straits song, yesterday QOTSA, the other day some psytrance. So Spotify recommends all genres in the same playlist, creating an awful cacophony of songs that don't flow into one another. Spotify recommendation system is very bad in cases like mine apparently. It's so bad I keep listening to the same old songs instead of discovering something new, which is why I'm no longer a customer until they improve their AI.

breakfastduck 5y ago
I listen to a LOT of different genres of music and their Discover reccomendations (not the playlists, the actual you like X artist you might like X) are absolutely brilliant. Got to be approx 50-60% hit rate. It seems to differ wildly between people - I've heard plenty of people making the same case as you, but many that have the same as me. It never appears to be 'its ok', its either 'brilliant' or 'terrible'. I've been using Spotify for 10+ years at this point, so maybe that makes a difference?
newsbinator 5y ago
Same here. I like Christopher Tin (e.g. Baba Yetu) and therefore for 6 years I've had no way to tell Spotify I don't like 8-bit video game music. No amount of skipping or "I don't like this artist" seems to convince Spotify to stop offering me variations of 8-bit Tetris themes or MULTIPLE 8-bit Cowboy Bebop Tank! theme variations per every single Discover Weekly. I'd be thrilled to even get a 2% hit rate on music I like. Seriously Spotify: two separate "Tank!" versions per Discover Weekly is too many.
- kortex 5y ago
  Man, the spotify algo makes no sense. The other week I was in a really specific mood, starting with Sonic 1/2 OST and moving into 8bit/chiptune/keygen. I probably listened to every rendition of Chemical Zone out there. Rest of the week, much the same. Nothing in Discover Weekly.
  - marginalia_nu 5y ago
    I think the relative "power" of a music genre is in relationship to its popularity compared to the popularity of what you usually listen to. Stands to reason that if you listen to rarer music (i.e. stuff the algorithm has less information about), then stuff that is more common (and the algorithm knows more about) will have a disproportionate effect on your recommendations. This seems to be a problem with this general type of recommendation algorithms in general. YouTube suffers from it as well.

breakfastduck 5y ago

I listen to a LOT of different genres of music and their Discover reccomendations (not the playlists, the actual you like X artist you might like X) are absolutely brilliant. Got to be approx 50-60% hit rate. It seems to differ wildly between people - I've heard plenty of people making the same case as you, but many that have the same as me. It never appears to be 'its ok', its either 'brilliant' or 'terrible'. I've been using Spotify for 10+ years at this point, so maybe that makes a difference?

newsbinator 5y ago

Same here. I like Christopher Tin (e.g. Baba Yetu) and therefore for 6 years I've had no way to tell Spotify I don't like 8-bit video game music. No amount of skipping or "I don't like this artist" seems to convince Spotify to stop offering me variations of 8-bit Tetris themes or MULTIPLE 8-bit Cowboy Bebop Tank! theme variations per every single Discover Weekly. I'd be thrilled to even get a 2% hit rate on music I like. Seriously Spotify: two separate "Tank!" versions per Discover Weekly is too many.

kortex 5y ago
Man, the spotify algo makes no sense. The other week I was in a really specific mood, starting with Sonic 1/2 OST and moving into 8bit/chiptune/keygen. I probably listened to every rendition of Chemical Zone out there. Rest of the week, much the same. Nothing in Discover Weekly.
- marginalia_nu 5y ago
  I think the relative "power" of a music genre is in relationship to its popularity compared to the popularity of what you usually listen to. Stands to reason that if you listen to rarer music (i.e. stuff the algorithm has less information about), then stuff that is more common (and the algorithm knows more about) will have a disproportionate effect on your recommendations. This seems to be a problem with this general type of recommendation algorithms in general. YouTube suffers from it as well.