Spotify algorithms?

I hear that Spotify uses technology to learn what you like and gives really good music recommendations. I haven't used Spotify lately but I am intrigued by this. A couple questions:
1. Is it all bullshit? Like, is it any different from how any other streaming platform recommends music?
2. How are these recommendations laid out? Into albums, or a playlist, or what?
3. Have you legitimately found new music you love you wouldn't have found without Spotify?
4. How do these algorithms work if it's not bullshit?

Attached: Spotify.png (512x512, 39K)

Other urls found in this thread:

en.wikipedia.org/wiki/K-nearest_neighbors_algorithm
en.wikipedia.org/wiki/Logistic_regression
en.wikipedia.org/wiki/Random_forest
twitter.com/AnonBabble

its more for notmies than ppl who already know how to search out music they like

1. i dont know about other platforms. spotify is good at reccing music based on playlists for example. if i make a playlist with a couple of songs it will rec me songs to add to that (and these usually fit quite well.
2. as above, playlist suggestions, also you have your weekly mixes (at least 5 i believe, sorted by your favorite genres. so lets say i listen to lots of future bass, the first weekly mix will be future bass artists.
3. yes, lots of it.
4. they collect data from people that save songs or put them in their playlists (maybe also by counting listening time). i guess they then make taste profiles of peoples tastes and compare them to you. sorta like "if you liked this, youll probably like that, since x amount of users also like that band you like.

Yeah? How do you find music then?

*also theres the song/artist radios. you can go on a song's "radio" and it will infinetely play music similar to that song/artist.

Yes it works extremely well. You get a "discover weekly" playlist every monday that gives you like 50 tracks it think you will like. You also get more genre specific recommendations on the discover page. You also can get reccs based on any playlist you make. Just make a playlist and it will recc songs below the playlist that it thinks fit.
false. it recommends extremely obscure stuff as long as it fits what you listen to. I almost never get recommendations of major artists even in the genres that all my recs are for. I;ve found probably hundreds of artists through various Spotify algo methods over the last few years. It's not as good for obscure stuff in older genres, e.g. I never find any 70's prog I haven't heard through there because none of those bands have been together in 30 years so their stuff isn't uploaded, but for more current genres it's really good.

So will these algorithms work if I just press the thumbs up buttons on songs and albums? I'm not much into making playlists but I like the idea of artificial intelligence making one for me

What genres are you talking about here? I tried using it and it threw only very surface level stuff or genres I've never listened to at me.

It works by analysing the music you listen to a lot then making weekly playlists for you. So just listen to your music

Okay, thanks user

Do you share your account with a normie?

I get tons of good and obscure recs in ambient, arty math/ambient/semi-electronic/electro-acoustic rock type stuff, death metal, doom metal, techno, IDM/unchategorizable electronic stuff, modern classical, modern jazz / jazz/electronic crossover stuff, all kinds of stuff really. But like I said, genres with more current releases have more stuff on because the artists have to upload the work obviously.
Yeah, just hit the heart/save/favorite (they always change the name) on stuff you like. Just start by favoriting a bunch of albums from your favorite artists.

I don't use Spotify anymore but no, I used my account alone when I did
It didn't do shit for me in metal or punk, I either got Slayer recommended to me weekly or metalcore which I never listen to.

it works pretty well without you doing anything at all except using Spotify normally to listen to stuff you like - I don't bother with the likes and dislikes on the automated playlists and they are consistently pretty good for me

the Daily Mixes and the new releases for you -list are almost exclusively artists you've already listened to, so not great for finding new stuff but I think the 30 track Your Discover Weekly is pretty good as far as AI curated recommendation algorithms go

at the end of last year they generated a Tastebreakers playlist that consisted of 50 tracks from genres and artists you don't normally listen to that was really good for me

I honestly found some good albums through Spotify's algorithm

I'm seeing it really depend a lot on how much you listen to some genre - I listen to mostly jazz and the jazz recommendations are pretty deep, but for other genres I get more surface level stuff like for example Grateful Dead, Gentle Giant and Yes and Brian Wilson Presents Smile on this weeks list as the only things in those genres (to be fair all except Wilson were from less well known albums so I wasn't really familiar with any of those tracks, so in that sense they were still "new to me")

continuing on a little tangent, the biggest disappointment is actually regularly getting tracks I've listened to a ton off before Spotify came around and not really since then, so it's actually pretty well aware of what you've been listening to using Spofity, but when you start using Spotify you can probably expect to first get recs for stuff you're already listening to until the algorithm catches up to your listening habits with longer use, so don't judge it too much on the first pile of recs it throws at you

>How do these algorithms work if it's not bullshit?
en.wikipedia.org/wiki/K-nearest_neighbors_algorithm
en.wikipedia.org/wiki/Logistic_regression
en.wikipedia.org/wiki/Random_forest

it's definitely given me some music I wouldnt have found on my own. Ex. Listen to a bunch of doom, madvillian, etc, and it gave me recommendation for dabrye. Low and behold new doom track on his album I've never heard of that I play all the time now.

based

I try listening to artists I know I won't like to "broaden my horizons" and understand other people who like them. This has messed up the discovery tool and now it's mostly shit and a few gems that more often turn out to be duds. It's all my fault and I've given up using it.

1. It's not bullshit

2. The way spotify works is each new new account that's created is seen by them as a 'blanc profile'. Whenever you listen to a song, spotify registers that to your profile, and starts seeking out recommendations. The more you listen to a song, or artist, the higher they will rank them.

Let's say you make a playlist with only 2000's indie rock you listen a few times a week to, spotify will compare the songs in that playlist with similar user-created playlists, and start comparing which songs are in there and which aren't. The ones that aren't in yours, but are in alot of other similiar playlists will come to you in your daily mix, or underneath the playlists in 'recommended songs'.

My example, I've made a small Yea Forumscore playlist with some radiohead, sufjan, mbv etc. It recommended me other 'famous' Yea Forumscore songs that are often discussed here. This happened because other people from here have made similar playlists to mine, and spotify picks that up. In the end it's all based on a ranking system. There is a lot you can find about it online. Also, if you add your friends, spotify will rank them too and tries figuring out what kind of music you like because of him/her.

The front page and all it's different, that's all payola and stuff. When drake released his album every single popular playlist had his face on it, even unrelated ones, see pic related. Avoid those playlists is my suggestion

3. I'd say yes, but not that often. Mostly from the 'radio' that plays after you finish listening to an album or playlist. Discover weekly etc are shite in my opinion. Also: try obscurifymusic.com. It takes your data from spotify and gives you (better) recommendations, and also some interesting statistics.

4. see 2. But it's also a matter how susceptible you are to their algorithms combined with their marketing. I take my regular music recommendations from way other places. Music I randomly discover on spotify I see as a bonus

Attached: payola.png (1384x2335, 3.07M)

No it's more than likely simply CF based on matrix factorization (ALS or sth).

Okay so I just ran into a problem. I was listening to one of my daily mixes, and then I decided to get premium. Got premium, but now there's no longer a heart button, but a plus button. Just because I like a song doesn't mean I want to add it to a playlist. What gives? Why did it change?

on principle heart and nope should always be there for auto-generated playlists like Daily mixes, Your Discover Weekly and the new releases for you list , while the + button is there for static playlists that usually someone has actually created

the kinda weird exception is that artist and track radios have the plus -sign even though you'd likely assume that the artist radios are dynamically created and could be tuned to personal preference

so, short answer "what buttons you see depends on what kind of playlist you're listening to"

my discover weekly is absolute garbage, ive heard 95% of the stuff it recommends

on the other hand spotify has extremely good "similar artists" recommendations whenever you listen to an artist you like. you're quite likely to like the similar artists as well