The Dynamics of Music Exploration on Spotify | Hacker News Reader

The Dynamics of Music Exploration on Spotify

research.atspotify.com

For all the complexity... this is something Spotify has got right.

I'm in my 40s, and I still listen to an incredible amount of new music. That is across everything from mainstream rock and pop, through to jazz, jungle, soul, techno, house, hip hop, classical... everything (though I tend to draw the line at screamcore).

And in there last year a tiny unsigned band with 2 tracks and fewer than 900 listens (at the time) got suggested. That band is now one of my favourite bands, a Scottish punk band... Dead Pony https://open.spotify.com/artist/3lGO6uBIzoFyU6OoGMER1B?si=OT... . Since hearing them I've told my old contacts in the music business about them (typically DJs at 6 Music, a few people that own labels, and a few musicians too). Earlier this year they've been signed (nothing to do with me, though I hope whispers meandered and it helped), been featured on BBC 6 Music, and are now working on an album and one of their tracks has hit over 1M listens. I've seen them live and holy crap, they rock... destined for far larger venues than the 50 person under-arches theatre in London that I saw them...

Spotify is doing something very right when I'm not solely rediscovering things I've been listening to for 30 years, but am finding the things I'm going to listen to for the next 30 years.

I've tried Tidal, Deezer, Apple, and my own heavily loaded Plex with all of the music labelled and richly tagged... but the discovery capability of Spotify along with the playlists are their killer feature.

My experience is completely different. The weekly dynamic playlists are very poor (either completely the wrong genre or songs I regularly listen to in the case of the monday playlist). The daily playlists regularly play the same songs by an artist. It seems to me that spotify only uses the top 2 or 3 songs and nothing else. I'd say that of a specific genre it will play 40 to 50 songs and then repeat them.

Even randomizing your existing playlist plays the same music over and misses most of them. Not to mention the UI and flows are pretty terrible for a software that dominates the space.

Same, the only way I can reliably find new music I like on Spotify is to find compilation albums that artists I know feature on and listen to the other tracks.

Then a couple of years later half the new music I found is no longer on Spotify. I've largely given up on it.

This might possibly be true for the types of music you named. Try it with world music in general (e.g,. south asian or middle eastern music) and you'll come across severe limitations and suggestions that simply do not work. Spotify doesn't even still carry even half the catalog of music from some of these regions. It's suggestions in (western) classical also have left a lot to be desired.

which platform does this right though? YouTube is alright for Bollywood at least

The premise (i.e., a singular 'music streaming' platform is what one should 'judge' Spotify against) is itself what I would reject in this case. e.g, for Indian classical, you're much better off with a whole constellation of blogs out there (and separately, certain music festival sites that host their recordings). Even national radio works a lot, lot better.

The reason I run my Plex is all the world music, classical, and frankly obscure music.

If Spotify could increase and sustain the catalogue then I'm pretty sure this would work. But it feels like the catalogue is smaller, churns more, and isn't as enriched in meta-data or listener behaviour to make it work yet. A network problem more than a technical one.

Yes, agreed, it's definitely a much bigger network & economic problem than a technical one.

Blogs, discogs and national radio actually work a lot better for me for discovery, simply because, as you say, Spotify may not even have the input data in order to then help me discover better music. Thing is, I'm not sure there is a desire (perhaps economically driven) to improve that catalogue as they would need to.

When I was at Spotify 3-4 years ago, one of the problems was poor metadata for most music. There's a semi-internal project to be better at classification (https://everynoise.com/) but it has its own problems.

And you can't rely on automatic classifications either. But then music providers don't give you proper info for the music they provide. It's a circular problem.

And then it's a social problem, too. You may want to listen to music that other people that listen to your music do... But then you might not want to do that.

It's a mess.

Also, the ability to listen to whole albums or discographies at a time when no one else was really offering that.

Part of what drove me away from Pandora at a certain point (maybe they've changed since then, don't know) was that they would only play a small subsection of any particular artist's work-- that is no bueno for me, if I really love them I want to hear it all at least once.

Gosh, I had basically the same story. Heard Dead Pony for the first time on some random Spotify playlist about a year ago, and they're awesome. I'm hopeful they do some shows in the US soon.

My experience with Spotify is similar to yours, but sometimes I want to go on a discovery binge... For that I use Every Noise at Once, built by a Spotify engineer (glenn mcdonald) and one of the coolest sites I've ever used:

https://everynoise.com/

The study is interesting, however I think Spotify's algoritms and user interface in combination with its promotional tools for publishers and labels are interfering with the results. These are specifically designed to make the users make certain choices. I still get loads of recommandations on music I never will listen to in the app. These are not strange and obscure music suggestions, neither are they promotion of artists and music within the broad spectrum of genres I listen to. They are in large promotion of the already top promoted artists in the world. Their search is not very good for finding new and interesting music either. In my experience you need to have very specific knowledge of what to look for, to get good results. So, when Spotify do this kind of research I would be very interested in how they take into consideration the effect their own activity have on their users behaviour? In my personal experience they constantly try to make me less explorational. I prefer Last.fm for exploring.

Besides the algorithmic recommendations, there's a big and increasing amount of Payola on Spotify (as well as the streaming services from Apple, Amazon, etc.)

Payola in the digital world is not blatantly done like in radio where you invite the DJs to a brothel and / or pay them some money, but by encouraging Spotify to recommend a certain artist / song more than others because it's supplied by a "preferred" partner.

How do you become a "preferred partner"? Well...take a wild guess.

The amount of data Spotify has on user behavior must be awe-inducing. Not that I'm complaining, the Algorithm has shown me plenty of good music that I would have missed otherwise. Seeing the data analyzed like this and the results published publicly is pretty cool!

I just wish sometimes there would be better ways to feed Spotify data on my preferences. I have a running playlist for example that I listen to over and over -- that does not mean I enjoy the genre 'DnB at 175BPM' so much that I want the same kind of music in my Discover Weekly.

It would be great if there either was a 'mood' you could set to say: I'm now listening to stuff that's only relevant in this context and shouldn't pollute other 'moods'; or if you could at least mark playlists as 'do not track'. I suspect you could already hide individual listening sessions from the data collection by making it a private session (would make sense at least, though I haven't checked), but who would remember to do that consistently, especially given how you need to set that switch manually in the settings.

+1, had to switch to another app for white noise to sleep at night after my discover weekly became exclusively color noise (https://en.wikipedia.org/wiki/Colors_of_noise)

That's hilarious, but also highlights a good point in our lack of control/observability into these algorithms

The idea is that consumers don't know themselves as well as they could - you could argue either way.

FYI you can mark sessions as private and the data won't flow into recommendations

I discovered this feature by random. Probably they should make it more clear.

I think anyone who has DM'd a game of D&D and used Spotify knows the pain of having your Discover Weekly taken over by "Epic Battlegrounds - The Drums of War 16: Dragons in the Sky"

For music discovery, nothing ever really rocked my socks like Pandora in its early days. I discovered many obscure bands with it - the software really lived to its promise of finding you tunes with similar musical characteristics. Almost two decades later, Spotify's radio feature is very lacklustre in comparison. It seems to be heavily based on social signals and not enough on the features of the underlying songs. I can and do discover music with Spotify but pace is very slow, and I have the feeling that things could be so much better.

Pandora, the good old days. Using some USA proxy to get it in Europe and using some obscure downloader to save the best songs before music streaming was the thing.

The recommendations where perfect! For any genre.

It was a solved problem, Spotify sucks really bad at recommending things. Weekly playlist are rubbishy and the radio feature is also not really good.

I used it from Spain and I don't remember even needing a proxy in its earliest days. I think the IP issues and fencing came later... I've considered trying it nowadays but I think one also needs a US card - way too much friction. I'm curious if it's still a good tool for music discovery.

There's an entire industry of pluggers and marketers doing whatever they can to get tracks on those Spotify playlists. And also businesses which link artists to playlist curators - for a fee, of course.

There's no sense in which they're an organic reflection of anything at all.

Pandora was never big enough to suffer much from that problem.

https://www.pandora.com/about/mgp

Pandora is still there and still pretty good!

Not where I live :(

Glad to hear it’s doing well. Back in the day it felt like one of the most impressive pieces of software I had ever used.

I am a little bit too invested in music listening and Spotify pretty much killed my appetite for looking out for new music in their UI. So it became a Drive for my MP3s.

Whatever Spotify is doing is not coming close to my experience with Pandora while it was open internationally, it was seriously ridiculously good. LastFM was also nice though not as good but definitely much better than Spotify. The issue is probably with record companies pushing songs that are not necessarily nice but promoted and squeezed in any chain that involves some relevant songs.

These days, I'm busy with downvoting the same songs hoping that the machine would finally get it that I don't like it. You would be surprised how many times can Steven Wilson can appear in my playlists. Porcupine Tree was nice, I liked it...fine but I don't want his other projects. Spotify disagrees so everyday I have to skip 20 songs from him. Everyday! You can try it yourself, try any progressive playlist and you will be bombarded.

Even Shazam was offering better stuff in their tiny box when I use their app before apple intervened.

Spotify is still one of my top ways to discover music, but I needed even more.

I’ve been spending a lot of time and money on discogs lately. I’ve found so much stuff you simply cannot find on the open web, Spotify, YouTube, etc.

Totally agree about discogs. For music outside the western world in particular, discogs is a godsend.

> Exploration spikes around Christmas, for example, during which people may be influenced to seek out season-specific music.

For my case, exploration does spike around Christmas, but this is due to all the "best albums of the year" lists coming out.

I get more than my fill of Christmas music involuntarily, no need to "discover" more.

First of all, what does “variety” even mean?

There used to be this cool script called "super eclectic score"[1] for last.fm. It had a pretty clever way of measuring how varied your tastes in music are.

The following script takes the 50 top artists in your musical profile from Last.fm, and finds the collection of top 20 similar artists (where the artist itself is the #1 most similar artist) for this top 50. The resulting is a list of artists similar to your preferred artists. As the list is larger (maximum = 1000), your musical preference is more diverse.

[1] http://anthony.liekens.net/pub/scripts/last.fm/supereclectic...

To be honest I use Spotify the same way I used iTunes back when I was a teenager. The main difference is that I'm paying to browse some company's filesystem instead of using my own.

Yes, this is how I use it, too. But we're not like the majority of users who use it like Radio - just picking a popular playlist or "radio" from the Spotify start page, and letting it play for hours on end.

That's why the business for the suppliers of Spotify has been reduced to basically pitch the songs they're trying to promote to the most frequented playlists.

That's how "payola" has made it into the digital realm. The algorithmic reco's on Spotify are a different, less interesting story for artists and labels.

> Our results are consistent with younger people being generalist repeat consumers who do not often look for very novel music compared to what they know, but the music that they do know is relatively broad. Older listeners are specialist explorers, who look for a specific and potentially ossified set of music but are also constantly discovering and turning content over.

I'm not a data scientist but I find it very hard to look at all those charts, many with a very narrow y range, and come to such a black and white conclusion.

Sure, the researchers might be pressured to extract "actionable" categorisations in order to justify their salary, but since they released it to the public, you bet we're going to see hundreds of "young people are X, old people are Y" articles citing this study.

Why can't they say "a larger percentage of younger listeners (...)"? Argh, I guess that's why I grew to distrust the whole social and medical sciences fields.

Great blog post, appreciate the level of detail on the results and metrics looked at. Would be very curious to see how genre impacts things and if age groups remain stratified (or become more stratified?) controlling for it.

Is this link broken? Seems like the page is down.

Reply on news.ycombinator.com