[runj__]

Having talked with people that have close insight into Discover Weekly it actually only checks your last few thousand songs though (weighting them accordingly) and checks if those songs are on other peoples playlists, then choosing the songs on their playlists that are not on yours. This means that you have a pretty big pool of songs to start with (people with similar taste) that is getting smaller and smaller. The fact that those people also use discover weekly could lead to a bit of a feedback loop.

I found a friend who during a party came up and said "I listen to this song all the time, it's amazing!" and then the next and the next and the next song the same happened. We had been listening to the same (semi-obscure) music for the majority of the past year. It was only later that we discovered that we had both found the songs from discover weekly and were not at all as obscure and knowledgable about music, but hey: we had the same taste in music.

[gcr]

Please don't reveal intellectual property secrets on HN. It sounds like you got that information from a privileged source.

(EDIT: I was wrong; this algorithm is public knowledge.)

[karim]

Ha, those algorithms are far from trade secrets! Here's one of the Discover Weekly developers talking about how they implemented this feature: https://atscaleconference.com/videos/creating-and-scaling-sp.... If you're interested, here's a deeper technical explanation of the algorithm used: http://www.columbia.edu/~jwp2128/Teaching/W4721/papers/ieeec...

[gcr]

I see. Pardon, my mistake.

[mikeash]

If they didn't tell this person not to share the info, then the problem rests with whoever divulged this information in the first place. Once it's out, it's out.

[hbosch]

It is common knowledge[0] that Spotify peruses and pilfers songs from users with similar listener profiles and trades them around under "Discover Weekly". There aren't any closely guarded IP secrets in the GP comment here.

-

0. http://qz.com/571007/the-magic-that-makes-spotifys-discover-...