How Spotify Uses Big Data and AI to Pick Your Next Song | NYCDA Blog

Going in I had a few select questions, mainly why my #1 song on my 2016 Year in Review was “Venus” by Lady Gaga, despite all the times I set it to Private Mode for my ritualistic guilty pleasure listening sessions.

More importantly than that, I came to learn how big data and machine learning are used by the engineers over at Spotify.

With over 100M registered users worldwide, Spotify has many unique data and AI challenges. There’s 25 billion data points within Spotify’s database, 100 million music preferences to catalog and cater to, over 30M songs, and 2 billion playlists across 60 countries.

In the past, data logs at Spotify were put in by interns or those on the lower tiers of the company. Due to the extreme demand, errors were made—which in the world of stream revenue, one screw-up can be costly to artists, labels, and investors in the music industry.

The solution to this is AI—and the big data that the AI has been gathering over the years makes it a lot smarter in terms of recommending songs. The more user data fed into Spotify, the better at predicting future songs functions like Discover and Discover Weekly get. By just using one 1 song, the machines over at Spotify can gain 4 key points from users: metadata, interactions generated, who users are, and playlists.

This results in all songs operating within Spotify's network—what's known as an Audio Vector Space.

Spotify’s machine learning comprises of two main functions. The first is its deep learning network that’s made up of simpler learning ‘nodes’ that learn from the user—whether they skip or stay on a particular song, how frequently they play particular songs, among other things.

The second is a “convolutional” neural network, which is made up of cover art, layering songs on songs through matrixes to find similarities in lyrics, tones, frequencies, volume, and what mood it conveys. This network refines its knowledge based on the trial and error from the deep learning network, but layers it over and over to the point your Discover and Discover Weekly pages speak directly to you.

Listening to music is one of the most intimate and personal human experiences you can have. So much is packed into how and why we listen to any given song: whether it brings back good memories from yore, reminds you of an ex, or songs that match your mood whether it’s channeling your inner diva in the mirror to Beyonce or ugly-crying in the bathroom to Adele.

The biggest objective coming from Spotify is recognizing how a music session represents a “moment” for the user and using data to describe the moment. Through its machine learning, code helps break down the emotional aspect into a series of systems that can, over time, improve its own tastemaking abilities.

In case you've ever had the thought "dang Spotify really knows me!" it's because of your entire usage history.