This is an excellent overview of how the team at Spotify conceived of and brought to life Discover Weekly, their algorithmically generated, personalized playlists of uncannily accurate recommendations for each user. It touches on a few data science concepts that will escape many readers (like myself) but overall it provides great insight into what is, for my money, the best recommendation technology I’ve ever seen (or heard). The story of how this feature evolved from previous attempts at greasing user discovery of content is on its own very eye-opening, as is the revelation that repurposing users’ Facebook avatars as the “cover image” for the playlists led to a 10% increase in weekly average users.
The slides here are embedded from SlideShare which can be flaky, so if this doesn’t work for you, you can see the full deck at slideshare.net. Thanks to @adstyles for the tip.
+