4 ms·
Who is listening to all of this self-published music to get training data for Discover Weekly?
by whatok 9y ago
Who is listening to all of this self-published music to get training data for Discover Weekly?
- samfriedman 9y agoI'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.
- whatok 9y agoAgreed 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.
- noxToken 9y agoThey'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.
- kevin_thibedeau 9y agoYou 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.
- whatok 9y agoThis 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.
- maherbeg 9y agoI 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.
- lavezzi 9y agoIn-house positions or they scrape/steal from the blogs.
- whatok 9y agoDefinitely 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.
- skinnymuch 9y agoDiscover 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.
- heartbreak 9y agoI 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.
- ghaff 9y agoI'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.