Listen to this story

--:--

--:--

How Does Spotify Know You So Well?

A software engineer explains the science behind personalized music recommendations

Photo by studioEAST/Getty Images
A Spotify Discover Weekly playlist — specifically, mine.

A Brief History of Online Music Curation

Spotify’s Three Types of Recommendation Models

Image source: Ever Wonder How Spotify Discover Weekly Works? Data Science, via Galvanize.

Recommendation Model #1: Collaborative Filtering

Image source: Collaborative Filtering at Spotify, by Erik Bernhardsson, ex-Spotify.
Some complicated math…
The User/Song matrix produces two types of vectors: user vectors and song vectors. Image source: From Idea to Execution: Spotify’s Discover Weekly, by Chris Johnson, ex-Spotify.

Recommendation Model #2: Natural Language Processing (NLP)

“Cultural vectors” or “top terms,” as used by the Echo Nest. Image source: How music recommendation works — and doesn’t work, by Brian Whitman, co-founder of The Echo Nest.

Recommendation Model #3: Raw Audio Models

Sophia, we already have so much data from the first two models! Why do we need to analyze the audio itself, too?

Image source: Recommending music on Spotify with deep learning, Sander Dieleman.
Image source: Tristan Jehan & David DesRoches, via The Echo Nest.

Warrior for authenticity. Uncovering my truest self & documenting the journey. http://sophiaciocca.com