Personalized playlists, artist radios etc. The point is you don't have to sift because they do it for you.

Sort of, but not really. In reality a lot of the streaming contracts with record labels dictate that they must promote certain artists more than others.

Spotify playlists and Browse are very bias. My Spotify 'home page' shows me mostly popular/mainstream stuff that I never listen to. :)

It was my understanding that Spotify's home page generally shows global trends and probably sponsored content, at least it's pretty obvious for me because as you point out it's filled with artists and music genres I never listen to.

There's however plenty of automatically generated customized playlists available next to that. When you create a playlist it also recommends tracks it thinks will fit etc... I haven't found any obvious bias in those so far and they often recommend obscure old tracks that I doubt anybody cares to promote anymore.

Social playlists are where it's at. Expect Spotify to further leverage social and friend features.

The GP's post was implying that it's easier to bypass traditional gatekeepers (labels) by self-publishing on Spotify. I don't see how either of your recommendations solves the problem of the flood of self-published releases in a "labels are no more" scenario.

Finding music you like among all the competing possibilities is Spotify's strong point.

This is roughly the youtube model now, isn't it? Some grass-roots promotion will get content fed into a recommendation engine of sorts.

I built https://www.jqbx.fm to help with that. It let's other people play music for you. So if you know someone's whose taste or playlist you like you can listen to music in sync with him/her/them. Like a user-generated radio.

Discover Weekly

Who is listening to all of this self-published music to get training data for Discover Weekly?

I'd imagine that the self-published music gets discovered by a few people and put into playlists, then more people who follow those playlists discover those songs, and then they percolate upwards through layers of algorithmic and editorial Spotify playlist generation including Discover Weekly.

There's also the Fresh Finds playlists, which work a little differently by surfacing unknown songs that "tastemakers" have been listening to.

Agreed that your hypothesis makes sense given the current state of things but I do not think it scales with any sort of consistency/quality control in the utopian "no labels all self-published" world.

You algorithmically seed new music to those with an interest in the genre and keep the songs with higher replays around for more recommendations. With perfect knowledge of who is listening to what and their overall tastes you can cut out all the inefficiency of the current gatekeepers (labels, billboard, and corp. radio) who act as self imposed kingmakers for a few anointed acts. The cream will rise to the top naturally.

This sounds terrible from a user perspective and arguably promotes certain types of music while penalizing others. As a user, if I hear more than a few songs that are terrible that were forced upon me by some algorithm, I am not going to listen to that playlist/use that service anymore. If users are the only way you are generating training data (for this supposed only self-published music world), you are going to end up with the music version of clickbait to get users' attention before they hit skip. I really cannot emphasize how bad optimizing for this would be for music.

Whatever your opinion on "current gatekeepers", some of them do provide valuable filtering services that I struggle to see how an algorithm could accurately replicate without compromising what makes music magical.

I think you could do this by pushing to users who tend to look for new artists already. There's a large set of folks that love to discover new music, and actively do so, and adding one or two new artists to promote here would be an easy way to start the network effect to other users.

I don't know if Paul Lamere is on HN, but he's likely the best person to answer this question. Short answer: It's not all about user-fed listen data.

I've watched some of Paul's talks as well as seeing some of his early demos at Sun Labs. The bottom line is that discovery is really hard generally and, for the most part, no one has cracked it pretty much anywhere even when there aren't competing commercial interests.

They're not starting in a vacuum. Spotify has metric tons of listening and catalog data to help properly tag a newly signed artist. Discover weekly would help fine tune the data that's already on hand.

Discover Weekly as far as i know didn’t begin immediately or at least wasn’t what it is now at first. Now it’s great. Spotify has so much past and present data from listeners. Way more than they need.

In-house positions or they scrape/steal from the blogs.

Definitely would necessitate something resembling in-house positions but I just don't see how you maintain any sort of consistency if you have a group of people reviewing thousands of tracks. And then at that point aren't they just functioning as a label?

Blogs, don't know how they're going to discover stuff if it's all published on Spotify initially.