3 ms·
Pretty cool to see them go after the cold start problem with audio models. Often measuring the impact of having cold start vs. not is very difficult because it
by dontreact 9y ago
Pretty cool to see them go after the cold start problem with audio models. Often measuring the impact of having cold start vs. not is very difficult because it only affects the long tail of tracks at first, but eventually it could change the dynamics of discovery in the system.
- marcus_holmes 9y agoIt occurred to me, though, that this is wide open to hackery. Find a popular song that is likely to be on a lot of people's playlists. Make a new song that is a close match to it for raw audio modelling. Launch on Spotify. Yeah you'll probably only get one or two plays per user, but spread over a large number of users (who all listen to the whole Discover playlist every Monday, like I do) it's still significant traffic. Or have I got this wrong?
- dontreact 9y agoI imagine that once you get to any significant number of plays (N=1000, which would not be a significant payout on spotify at all), the collaborative filter model would take over and if none of the people who listened to it caused further positive signals as mentioned (visiting the artist page, repeat listens) then the track would stop being recommended pretty quickly. Also, what you are describing sounds like what producers already do because people actually like it: find the latest trends in sound and copy them, (but hopefully with a fresh twist so that people get into it) :)