How to get good recommendations on Spotify

[bluefuzz]

@Multicore - I've started a new thread because the other Spotify thread got locked for some reason.

@Multicore said:
I would appreciate a webinar or some such tutorial. I switched to using Spotify 6 months to a year ago from Amazon Music. I like the way it works with various control and streaming devices. That works great. But showing me music I don't know but like, I can't find it. Dunno where to look. Where is it? Show me, please!

Lol! I don't think I do anything special to get recommendations, nor am I privy to the arcana of Spotify's algorithm. All I know is what Spotify themselves say how their recommendations work. But I can describe how I use Spotify, and perhaps more importantly how I don't use it.

• As mentioned in the other thread, I use the Mac desktop client almost exclusively. In settings I have 'Autoplay' turned on and all the social features turned off as far as possible.
• I fairly religiously listen to the 'Discover Weekly' playlist every ...um... week.
• If a track piques my interest, I click the three-dot menu on the track and choose 'Go to album' and then click the big plus button at the top 'Save to your library', i.e. I save the whole album to the library. I (almost) never like/save single tracks. I also very rarely skip, hide or remove tracks from the recommendations or library.
• I never touch Podcasts

And that's about it! I really don't do anything else. I very rarely curate my own playlists. The only playlists I've made are if I have a bee in my bonnet over some song or musical rabbithole. For example I have a playlist of all the versions I can find of Antonio Soler's 'Fandango in Dm' (my favourite bonkers harpsichord piece) and another with a hundred or so versions of 'Danny Boy' (don't ask!).

I occasionally use Spotify's Made for You playlists or other user's curated playlists as a starting point if I want to deep dive into something, e.g. 60s Girl Groups or Musique Concrête. Again if a track appeals to me in some way I add the whole album the track is from to my library.

So when I want to actually listen to some music I go to my Library > Albums view and sort by recently added. There will typically be a fair number of albums in this view I haven't heard yet (other than a single track) so I may pick one of them or revisit some old favourite. Being an old fart I'm quite resolutely an album person. However, once the album is finished I usually let the 'Autoplay' feature continue playing which will usually be more tracks by the same artist, genre or similar 'mood'. If I'm not in the mood for more of the same then I just go and find something else to play.

I do 'follow' contemporary artists (there's little point in following dead ones) but have all notification features turned off so I don't know if that makes any difference. I do also occasionally use the 'song radio' and 'artist radio' features and again always add full albums to the library if anything good turns up.

I think having a fairly large number of albums in my library that I have liked but haven't necessarily heard or have forgotten about is key to my 'system'. The Discover and Autoplay recommendations are usually good enough that I never get recommended anything truly objectionable. It's rare that my Discover Weekly is devoid of anything noteworthy and quite often full of great tracks and I can add at least half a dozen new albums to my library.

Other advice is available ... ;-)

 

[Multicore]

gr8stuff. thank you for taking the trouble to type that.

i have a sore knee after an accident hiking with the dogs this morning so i may be on the couch quite a bit with ice compression and elevation. this should afford some opportunity to get started on the project.

i have now found "Discover Weekly". it's hidden behind block of icons called "Made For". if i click the little "Show all" text over the other side of that it's in there. i'll have a go.

i don't like auto play and turned it off. if i listen to the ciaccona from js bach violin partita 2 or a zorn hermtic organ improv then i want that to sit when it's done. i don't want to have to intervene with spotify's idea of what should immediately follow that. especially if spotify's idea of what i like is

yes on the marc ribot. but otherwise ... seriously? where did it get that from?

i'll get started with "Discover Weekly" and report back.

 

[bluefuzz]

i don't like auto play and turned it off.

Yes, I get that. It can be a bit annoying, but I'm lazy and just leave it on. Sometimes the recommendations are so similar to what I just listened to that I don't notice it's a new artist ...

yes on the marc ribot. but otherwise ... seriously? where did it get that from?

Hmm. Those last three are also in my 'New Releases' view. However, above them are new releases from Japanese Television, Stereolab, Hakushi Hasegawa, Kinloch Nelson, Pye Corner Audio, Bernard Parmegiani, Mary Halvorson, Tinyhawk & Bizzaro and Jethro Tull (!) all of whom are probably in my library somewhere. At least they have the decency to put the payola stuff at the bottom. The Ribot is already in my library: a bit disappointing though. I listen to Ribot for the guitar playing not his singing which is simply uninteresting and mildly annoying. And the songs are pretty weak on this one too.

I recently downloaded my complete listening history from Spotify – I've listened to about 70,000 tracks in 10 years! – and connected it to Listenbrainz scrobbling service. They also have some autogenerating playlists that can be based on what you have listened to as well as what you haven't listened too. I think you can connect your Spotify account directly. The playlists can also be exported to Spotify. It's certainly a different way to slice and dice the recommendations.

Good luck with the knee ...

 

[Multicore]

I like some of the Ceramic Dog songs very much. B-Flat Ontology for example. He's not much of a singer but the singing provides form and the lyrics have some uses. But, sure, it doesn't always work.

A thousand singer songwriters each more earnest than the next strip of layers of pretension until there's nothing left.

Oh what will Zizek think of next? What will Latour and Zizek think of next?

 

[bluefuzz]

I like some of the Ceramic Dog songs very much.

Yes, the protest songs from a couple of years ago was pretty good too: there was some rightious anger there. But seriously, when you're pretty much the best guitarist on the planet, we don't need you to sing ...

 

[Brian Hall]

Like and add artists / music you like. Skip things you don't like. It takes a while, but then the recommendation algorithm works very well.

 

[Multicore]

Discover Weekly was a bust. 90% was ECM euthanasia soundtrack. Ug.

 

[Hayabusa]

search a song you like, select "song radio" for that song, like the songs you like, repeat for other songs.

 

[bluefuzz]

Discover Weekly was a bust. 90% was ECM euthanasia soundtrack. Ug.

Patience Grasshopper!

Recommendations are based on what you actually play from your library on Spotify. So if you don't have a largish library and haven't actually played much from it then recommendations will be off. When I first subscribed to Spotify I added a good few hundred of my favourite albums to the library. It might take a few weeks to 'get to know you'. Just keep listening to a broad variety of stuff, add/like anything and everything that interests you to the library, and recommendations will get better. Personally, I almost never use the 'hide' function but if there are genres or artists you never want to be recommended you could try that ...

 

[Brian Hall]

It took a while for Spotify to stop recommending rap crap and stuff like that which I would never listen to. Eventually it gave up and now the recommendations are really good. The same goes for Youtube Music.

Qobuz never improved. Same for Tidal. Amazon Music is better at recommendations than either of those, but maybe half as good as Spotify.

 

[mns3dhm]

Agree that Discover Weekly (updated every Monday) is good at identifying songs that are related to songs I’ve liked, added to a playlist, listened to, etc. I also look closely at songs mentioned by persons writing reviews of audio gear - I’ll usually head over to Spotify and look at the artist mentioned and their most popular songs. I spend some time looking at music reviews published by sites like All Music Guide, Rolling Stone, Spin, NME, Pitchfork, etc. If I read something that sounds interesting that I’m unfamiliar with, I’ll check out that artist on Spotify.

 

[bluefuzz]

It should go without saying (but I'll say it anyway) that the algorithm will never recommend something a million miles from anything you have listened to and liked, i.e. if you only listen to Country & Western and Military Marching Bands then it's not going to start recommending Mongolian Throat Singing or Finnish Black Metal unless you don't have enough history in which case it just defaults to the standard chart hits and payola tracks. So if you want varied and interesting recommendations you need a history of listening to varied and interesting music.

It should also perhaps be stated explicitly that only liking/adding albums to the library is alone not enough to get good recommendations. Nor is just listening to some tracks, e.g. from a playlist, without adding the playlist to your library or, as I do, whole albums to the the library. You have to listen to the tracks AND have them in the library for them to trigger recommendations.

 

[HalSF]

I'm intensely skeptical and hostile about Spotify's Discover Weekly offering a trustworthy and healthy route to new-music exploration. From what I've read it takes your listening history and uses AI and algorithms to given you more of the same while pushing you toward a crowd-sourced/groupthink stream of what other people are playing, plus an unknown quotient of Spotify spinning its own marketing and engagement agenda into your life. You may think you're gaming Spotify, but my guess is that Spotify is gaming you.

The AI behind Spotify's Discover Weekly

How does Spotify's Discover Weekly always nail your music taste? AI, machine learning, and data science work together to curate your perfect Monday playlist.

thewordharbor.com

"Spotify’s Discover Weekly isn’t just a lucky guess. It’s a carefully engineered blend of machine learning, massive data analysis, and some clever recommendation techniques. If you’ve ever wondered how Spotify seems to know your taste better than you do, it’s time to pull back the curtain on the tech that makes it all happen."

I'm not criticizing anyone else's choice — the impulse to find and support new music is a good thing! But my loyalty goes to direct human engagement and enthusiastic music reviewing and criticism, friends, and keeping the antenna up for solid honest leads on the cool newness, not what the robotic Swedish Borg is cooking up in the resistance-is-futile kitchen.

p.s. WTAF is an "ECM euthanasia soundtrack"?

 

[bluefuzz]

From what I've read

So you've never actually used it?

Of course they are using machine learning, AI, statistical techniques and whatever else can keep bums on seats. That's the whole point.

If you’ve ever wondered how Spotify seems to know your taste better than you do

While I've discovered a hell of a lot of great music through Spotify, I'm under no illusion that Spotify 'knows my taste better than I do'. Probably less than 5 % of its recommendations are truly on point. But that's better than nothing.

my loyalty goes to direct human engagement and enthusiastic music reviewing and criticism, friends, and keeping the antenna up

They are not mutually exclusive. Just because I use Spotify's algorithms doesn't mean I don't also find music elsewhere. My friends have crap taste.

You may think you're gaming Spotify, but my guess is that Spotify is gaming you.

I don't think I'm gaming anything. And how exactly is Spotify gaming me? Recommendations for long dead obscure African percussionists or equally obscure Scandinavian free jazz artists, who may get 50 plays a year, doesn't seem like a good strategy for world domination to me ...

 

[Multicore]

It should also perhaps be stated explicitly that only liking/adding albums to the library is alone not enough to get good recommendations. Nor is just listening to some tracks, e.g. from a playlist, without adding the playlist to your library or, as I do, whole albums to the the library. You have to listen to the tracks AND have them in the library for them to trigger recommendations.

You've mentioned this a few times now so I just want to clarify a technical point. If I listen to something for the first time while it's not in my library and then I add it to the library, is that alone enough to influence recommendation? I ask because you've phased it before as listen to something that is in your library, which suggests that the order needs to be the other way around, i.e. library first, then listen.

 

[bluefuzz]

If I listen to something for the first time while it's not in my library and then I add it to the library, is that alone enough to influence recommendation?

That's where I have to say I don't know. My intuition is that it doesn't matter. However, it would be weird if you have to add something to your library before listening to it for it to trigger a recommendation. Probably any and all interaction with the UI has some kind of statistical 'weight'. Adding tracks to your library and listening to them having (presumably) the highest weights. But I would think it is a simple addition where order doesn't matter ...

 

[Multicore]

"ECM euthanasia soundtrack"?

The music they play in the waiting room of the assisted suicide clinic that looks like it took design cues from Black Mirror.

 

[HalSF]

The music they play in the waiting room of the assisted suicide clinic that looks like it took design cues from Black Mirror.

Thanks. That seems like a really opaquely weird response to music released by ECM based on my experience with the label, but maybe Spotify is doing something ghastly and dark with it based on your listening history.

 

[Multicore]

Is there a keyboard shortcut for "Go back"?

My habits with web browsers make me press Command-LeftArrow, which interrupts playback. Very annoying.

 

[bluefuzz]

My habits with web browsers make me press Command-LeftArrow, which interrupts playback. Very annoying.

Yes, I do the same to much weeping and gnashing of teeth. The correct shortcut in the Spotify app is Alt + Arrow (left for back, right for forward) but I always forget until its too late ... ;-(

Here a few more shortcuts.