The lie of music discovery algorithms • Hacker News Board

Re: The lie of music discovery algorithms

#101

Post by fantasybuilder » Tue, Jul 30, 2024, 5:27 PM UTC

ratiolat wrote: post #22

Interesting. For example how would one find a "new" Tool or Deftones? Current algorithms probably don't "pick up" not-yet-so-popular things. For example Shelton San (I found out about them via word-of-mouth), although I'm frequent user of Spotify. This means that classical promotion channels are still necessary, as otherwise things get lost in noise.

I noticed that Spotify surfaces similar artists who are also of a similar popularity. So it's not like it doesn't understand that particular style, it just has to somehow pick a couple of dozen artists to show in that very coveted spot.

So what worked for me in the past is finding less popular artists and then checking their similar artists.

Re: The lie of music discovery algorithms

#102

Post by esafak » Tue, Jul 30, 2024, 5:36 PM UTC

Tostino wrote: post #41

khafra wrote: post #40

Earlier quoted context omitted.

If you have any taste at all other than "maximally dissimilar to anything I have liked before," there should be a feature that predicts songs you would like. If your taste is exactly "maximally dissimilar to anything I have liked before," that's actually pretty easy to calculate from the embeddings as well.

It didn't sound like op wanted maximally dissimilar from what they've liked before, but instead maximally dissimilar from what they've listened to recently.

Either way, a preference for novelty can be measured. They also let you edit your Taste Profile it seems: https://support.spotify.com/us/article/your-taste-profile/

Re: The lie of music discovery algorithms

#103

Post by MattGrommes » Tue, Jul 30, 2024, 5:44 PM UTC

drivebyhooting wrote: post #94

cush wrote: post #82

So I've been using Tidal for 5 years now, and feel they run circles around Spotify in terms of curation. Their algorithms and curated tracks are better, and they steer away from the social/gamification features and lean in to artist-centric features. For example, they've had a "credits" feature since day one - you can look up the producer, guitarist, oboeist, etc of any song, and see what other work they've done. In…

Oboeist? That is oddly specific. Why that example?

Some people like particular instruments. My son plays the bassoon and loves trying to look up who plays on a song if he hears one.

Re: The lie of music discovery algorithms

#104

Post by AlbertCory » Tue, Jul 30, 2024, 5:48 PM UTC

I'd be quite happy if Spotify just went away. The world of music would be much better off.

My solution (doesn't work for everyone): I have a large library on the microSD card on my phone, and set the music player to Shuffle. Quite often a song comes on and I think, "Wow, I own THAT??"

OK, I'll admit that doesn't play any new music. However, no bills for bandwidth!

Re: The lie of music discovery algorithms

#105

Post by munificent » Tue, Jul 30, 2024, 5:56 PM UTC

angry_moose wrote: post #85

The two issues I've had with every discovery algorithm: "We have [favorite band] at home" - it picks things you like from your favorite band - instruments, tempo, etc then finds bad knockoffs that are superficially similar but painful to listen to. The "Iron and Wine" problem - some bands are so generic that they tick every single similar box and flood your recommendations. For years, it didn't matter what band/genre…

I think a more fundamental problem is that people like music for very different reasons. Even the same person may define "similar" very differently at different points in time.

If I want music "like" "Groove is in the Heart", is it because:

* I want mid-tempo house-like dance music

* I want major key songs with female singing

* I want songs with rap interludes

* I want 90s music

* I want fun party music

* I want music that reminds of that awesome trip I took with my friends a few years ago where we played a bunch of songs over and over

There is no right answer to this question. But, outside of just looking for playlists, no music app I've seen gives you a way to specify in what way recommended music should similar to the current song.

I see this effect most acutely when I listen to something that happens to be popular. For many people "heard it a lot when doing this fun social thing" is one of the main reasons they like a particular song. This was true for me too when I was younger. But for me today, I'm mostly oblivious to popularity. I just like stuff that sounds a certain way.

Whenever I stumble onto a song that has a particular sound I like that happens to be well-known, the recommendation algorithm just starts throwing other popular stuff at me that sounds totally different.

Re: The lie of music discovery algorithms

#106

Post by esafak » Tue, Jul 30, 2024, 5:59 PM UTC

bsenftner wrote: post #10

Music discovery has never worked for me, for the simple reason it's the lyrics, not the music. I listen to anyone speaking truth, (the truth I believe, of course), and that gives me a wide disjoint range of music, but they are all singing about political social truths. Marvin Gaye, Public Enemy, Rage Against The Machine, Beyonce, The Stranglers, The Jam, Psychic TV... It's the lyrics, and now today, we finally have t…

I'm pretty certain Spotify uses lyric embeddings as an input to the recommender. They'd have a hard time recommending podcasts otherwise.

Re: The lie of music discovery algorithms

#107

Post by lowbloodsugar » Tue, Jul 30, 2024, 6:01 PM UTC

When I have people over, I hand them the iPad that runs the audio system (I use Roon, Qobuz and Tidal but I imagine anything will work). When they play new and interesting things, they are now in my history. My favorite discovery from this was Massive Attack's Teardrop, which was a track I had heard before (and loved) yet completely forgotten about years later.

Re: The lie of music discovery algorithms

#108

Post by vhiremath4 » Tue, Jul 30, 2024, 6:05 PM UTC

madmountaingoat wrote: post #88

diob wrote: post #57

Earlier quoted context omitted.

I'm not sure how the Spotify recommendation algorithm works at all, but for some reason I imagined them doing fancier things than looking at my liked songs and finding similar ones. I would've thought they'd build a profile of you, and then find similar user profiles and show you songs those folks liked that you hadn't found yet. That's gotta be how they do it, right? I'm probably wrong.

I don't work at Spotify anymore and I didn't work on the tech I'm describing, but I picked up a bit about what was going on while there. First, there is/was no single algorithm, but the core ideas driving a lot of recommendations is: 1. Create user taste vectors 2. Match those vectors to other users or collections of tracks 3. Use that information and combinations of other things to find recommendations. Each step of…

So collaborative filtering?

Re: The lie of music discovery algorithms

#109

Post by lowbloodsugar » Tue, Jul 30, 2024, 6:13 PM UTC

munificent wrote: post #105

angry_moose wrote: post #85

The two issues I've had with every discovery algorithm: "We have [favorite band] at home" - it picks things you like from your favorite band - instruments, tempo, etc then finds bad knockoffs that are superficially similar but painful to listen to. The "Iron and Wine" problem - some bands are so generic that they tick every single similar box and flood your recommendations. For years, it didn't matter what band/genre…

I think a more fundamental problem is that people like music for very different reasons. Even the same person may define "similar" very differently at different points in time. If I want music "like" "Groove is in the Heart", is it because: * I want mid-tempo house-like dance music * I want major key songs with female singing * I want songs with rap interludes * I want 90s music * I want fun party music * I want musi…

See, "Groove is in the Heart" makes me think of "Calling all units to broccolino" by Calibro 35. So I might add:

* Musicians having fun with instruments.

Re: The lie of music discovery algorithms

#110

Post by belthesar » Tue, Jul 30, 2024, 6:22 PM UTC

munificent wrote: post #105

angry_moose wrote: post #85

The two issues I've had with every discovery algorithm: "We have [favorite band] at home" - it picks things you like from your favorite band - instruments, tempo, etc then finds bad knockoffs that are superficially similar but painful to listen to. The "Iron and Wine" problem - some bands are so generic that they tick every single similar box and flood your recommendations. For years, it didn't matter what band/genre…

I think a more fundamental problem is that people like music for very different reasons. Even the same person may define "similar" very differently at different points in time. If I want music "like" "Groove is in the Heart", is it because: * I want mid-tempo house-like dance music * I want major key songs with female singing * I want songs with rap interludes * I want 90s music * I want fun party music * I want musi…

Ironically, this is something that I think Pandora solved quite well with their recommendation engine. By virtue of creating a station around a particular vibe, even if 5 playlists all started with the same seed song, weighting other songs up and down on each station would curate a different listening experience, by virtue of finding how those are similar. Where Pandora was limited (at least, the last time I used the service) was the pre-seeding process is a bit arduous and opaque. I'm not sure how you make that easy to interact with, as going a layer beneath to the "why" a song was recommended and allowing folks to influence the graph at that layer sounds like a daunting UX challenge.