The long tail is more alive than ever, but algorithms have always been a bad way to find it. Instead, listen to personal recommendations on web forums etc. There are people who are passionate about any obscure micro-genre you can imagine, and they'll point you to the best of it. Bandcamp and Soundcloud are full of long tail music that's completely ignored by the mainstream music industry.
It's interesting, I've gone back to Spotify from Apple Music primarily because I can easily find more interesting playlists from curators.
I don't think Spotify has any kind of special algorithmic sauce, they just make it easy for people create and share playlists.
It amazes me how most of the other streaming services try to turn everything into an algorithm, via "special radio" stations, etc. And yet, the best way to provide a long tail for music still seems to be basic web 2.0 - let the people make the content for you.
There's definitely some algorithmic shenanigans going on with Spotify on any of their auto playlists. Whatever algorithm they have decides that I absolutely love one specific artist and shoves them down my throat for months.
For a solid year, any time I played a Daily Mix or let one of my personal playlists runout and it reverted to "recommended music", Neko Case was always in the first top 3 songs. Before that it was Iron and Wine. I don't mind either artist, but I have never chosen to play either one myself.
I do like Spotify's curated playlists though. They are always a better source of new music than the automatic playlists.
Right, I'm not saying Spotify isn't trying out algorithms at all, but their algorithmic stuff doesn't seem much different from Apple Music, Google Play, etc. A lot of the "algorithmic" stuff enters into "lame" territory fast, where you see the issues like artist fixation, or "variations on a theme" where you get like 5 remixes or live cuts of the same song.
The main benefit of Spotify is that I can just bypass the algorithms, and go straight to people. And that's really where the "long tail" effect seems to thrive.
In my humble opinion all discovery algorithms with all large providers are terrible. Instead of expanding they narrow down choices and never surprise me in a good way like some human curated playlist can.
I had been switching between Apple Music, Spotify and Google Music in different countries. Currently I have settled on Google Music simply because Google practically abandoned Google Music so they are not so in my face with curated playlists and stations.
On the other hand I think that there is opportunity for good (like in really working) AI engine for music recommendations.
The unofficial gmusic API is why I continue using gmusic. I make many playlists for work (teaching spin classes) and I use the API to aggregate the songs I use and clean out old playlists.
If you let an algorithm script a movie, it would be a 2-hr superhero movie with extremely simplistic moral conundrums, that will cost $300m to make, but will gross $1 billion +.
If the algo scripted a TV sitcom, it would be 5-6 mostly white people who occasionally hook up, break up, get married, struggle to conceive, and then everything eventually works out.
Same goes for music. People have a habit of liking what's in front of them. Few are going to want to make the effort to find something that's more than "good enough".
As an anecdote I have left Apple Music after it had been suggesting to me at least 3 heavy metal playlists for 30 days. And this is the one genre I do not listen to.
I had been clicking 'I don't like it' buttons hoping the AI will get it but no luck. So after 30 days I had enough and moved on back to Spotify.
Maybe you actually like heavy metal but won’t admit it to yourself ;).
(Just kidding, but this kind of thing will be more common in the future, especially as we learn more about how little the brain knows about its own desires and motivations)
I don't know if this is (still?) true, but I've heard that user-generated playlists are their special algorithmic sauce, at least for "Discover Weekly". It's designed to take songs you're familiar with, look for them in other peoples' playlists, and fill the gaps between songs you already know about with songs that other people have filled the gaps with in their own user-generated playlists, but that you haven't heard yourself.
It sure seems like the a lot of suggestions on Spotify try to find "similarities" based on things you've listened to recently. Makes sense that the similarities are very likely pulled from recent popular additions ... from curators.
My sense is that is probably as good as it tends to get, unless you want to get incredibly creepy and start bringing outside personal information, e.g., associating that the 40-something living in Chicago may have an affinity towards Wilco. Personally, I'd rather deal with "meh" suggestions then have my music service track detailed demographic models of me.
> It amazes me how most of the other streaming services try to turn everything into an algorithm, via "special radio" stations,
I think they do that because it's the only way to keep people from switching to rival platforms. Spotify's selection isn't that different from Google Play Music/Apple Music/Amazon Prime Music etc. There's also little in the way of community interaction on those platforms beyond "sharing playlists".
SoundCloud at least has a system where you can comment on specific parts of a song. It creates a really interesting visualization of which parts of a single song people enjoyed the most.
