4 ms·
I'm genuinely confused by people who complain about Spotify's recommendation algorithm. I have eclectic, often obscure tastes, and I've found it to be an incred
by campground 5y ago
I'm genuinely confused by people who complain about Spotify's recommendation algorithm. I have eclectic, often obscure tastes, and I've found it to be an incredible way to discover new music.
I think you do have to prime the engine a bit. I "like" a lot of albums and songs. Every time I hear something new that I like I press the little heart button. Same when I discover music outside of Spotify (I listen to WFMU a lot). I make a lot of playlists. And I use the radio feature all the time. Click on the ellipses next to any song, artist, or album and you can "Go to radio" and Spotify will spin off a playlist of related music. I'm continually discovering new things that way.
My Discover Weekly and Release Radar playlists are always full of great stuff.
- ascorbic 5y agoI also find it strange. I find the algorithm incredibly good. I did link it with iTunes back in the day, so it started off knowing what I like, and I always Like tracks when I like them. Now I often find myself pressing like on most tracks on Discover Weekly because it's so good at recommending things I'll enjoy, and quite a lot in Daily Mix and Artist Radio too.
- hk__2 5y agoThe radio feature is awesome to discover new music. You can make a short playlist of a few songs you like, and start a radio from that. You can later tweak the playlist to change what you get in the radio while it’s playing.
- hkon 5y agoYes, same, I enjoyed the weekly recommendations for about a year, each week being recommended some songs which remain favorites. But the algorithmic recommendations then got heavily skewed towards Finnish rap music. Which made the recommendations garbage. The convergence towards Finnish rap is not uncommon on Spotify.
- oriolid 5y agoI've never had the problem with Finnish rap, but at one point the discover weekly was full of death metal. Also, for some reason they pushed really hard Ricky's Hand by Fad Gadget (awful, but at the time they had a well hidden ban button), Swamp Thing by Chameleons (kind of OK, but don't they have other songs?) and Love Will Save You by Swans (kind of OK, but do have long discography so why always that one song?)
- emerged 5y agoIt’s really good for me. But I had a large catalog of mp3 accumulated over the years and then manually followed a ton of those artists and liked their tracks. Good recommendations obviously require good data. You sort of get what you put in to an extent.
- joe_the_user 5y agoI'm genuinely confused by people who complain about Spotify's recommendation algorithm. I have eclectic, often obscure tastes, and I've found it to be an incredible way to discover new music. I too have "eclectic, obscure" tastes and I couldn't get anything from the recommendation algorithm and canceled after a month. I don't remember the thing recommending anything that seemed more obscure than what I searched for. And I made playlists. But seriously, some people's tastes are served by Spotify and some find it frustrating. It would be nice if those who don't get immediate satisfaction had a more sophisticated approach available. My hunch is the company would prefer less customers and more control.
- xvector 5y agoIt definitely took a couple of months before Spotify honed in on my tastes, but it was definitely worth the wait.
- Broken_Hippo 5y agoTo be fair, the recommendations are based on what you listened to lately in addition to what you've listened to in the past. A month probably wasn't long enough - it isn't really an immediate satisfaction thing when you have off-center tastes in music. If 70% of folks like popular music, that's the easiest thing to recommend. They might want a bit more control, I don't know, but I very highly doubt fewer customers is the goal. If they wanted fewer, they could just pull the free model completely. I'd guess piracy would increase and they'd get blamed, but I don't know.
- joe_the_user 5y agoTo be fair, the recommendations are based on what you listened to lately in addition to what you've listened to in the past So each day when my mood changes, I'm suppose start searching for a bunch of tunes and after an hour, I get what I want? If 70% of folks like popular music, that's the easiest thing to recommend. "Brilliant AI" ... turns into just top40 radio and victory is proclaimed. They might want a bit more control, I don't know, but I very highly doubt fewer customers is the goal This is what I said with a slightly different spin - they're willing to trade control for customers. Whenever companies engage in annoying or abusive policies that drive away customers, it's not that they don't want customers in the abstract, it's that they prefer control and profit margins over customers.
- 3np 5y agoI've been stuck in the same local optimum for years, getting the same songs and artists in the same genres (that I do and did like, but it's gotten old now)
- darthrupert 5y agoAgreed. A lot is wrong about Spotify, but the recommendations are superb.
- nikanj 5y agoSpotify's recommendation algorithm comes with a healthy pinch of Payola. "Hey you've been listening exclusively to chiptunes and power metal from the 80s. Here's a new single from $local_mumble_rapper you might love!" And you can't dislike the recommendation, so next week the algorithm is going to go "Oohh last week he played some mumble rap, let's give him a little bit more of it!"
- feffe 5y agoI did find it to be extremely poor but once I started liking individual songs, not whole albums it does better. Still, I'm not sure it uses other listening habits to feed the engine such as: - Early skipping a song from a generated play list (-1). - Repeat playing the same song from a generated play list (+1)
- hybridtupel 5y agoThe algorithm often is very strange. My mixtapes most often contain single tracks of audiobooks I listened to. Looks like Spotify treats audiobook tracks like any other song which is obviously very silly. Who doesn‘t want to listen to random audiobook snippets.