Hacker News Reader

> Pandora's discovery, recommendations, and shuffle algorithms are objectively superior

IIRC, these algos were part of a project called ~Muse which later forked, and there was a schism between the algos and clickstream data or the personal data and anonymized cohort data. Will have to review / re-search.

The trove of our searches, listens, and skips Pandora has / had did a good job informing recommendations and was an effective moat for my use.

View on HN

A big part of the algos was the "Music Genome Project"[1] as well. This allowed their recommendations to be about the music itself instead of just what other people also listened to, leading to some amazing music recommendations/discoveries.

[1] https://en.wikipedia.org/wiki/Music_Genome_Project

Thank you this helps -- the "Music Genome Project" was part of this fork & schism. Pandora says little about it[0], and the open portions[1] were last updated in 2014.

At least someone's still at it, and all this helps me find my notes from that time.

https://github.com/LDugar/MusicGenome/raw/master/Decoding-th...

[0]http://www.pandora.com/about/mgp [1]https://github.com/omgDB

My guessing is a lot of it went proprietary once competition started popping up trying to also make good music recommendations. "Secret sauce" and all...

Reply on news.ycombinator.com