The Evolution of Machine Learning at Spotify

Machine Learning allows for easier insights from massive data sets

Background

As of writing this, Spotify is the world’s largest audio entertainment platform, dominating the music streaming industry with a 31% market share (to Apple Music’s 15%) and rapidly growing in the podcasting space with a 20% market share (2nd to Apple Podcast’s 43%). To give a sense of the scale of Spotify here are some statistics:

· 489M Active Users

· 205M Paying Subscribers

· 100M Music Tracks

· 5M Podcast Titles

· 180 Markets

· 200 Petabytes of Content

Each of these 489 million users has a different personalized experience, having a different Home Page, with unique recommendations, liked songs, playlists, and podcast subscriptions. This massive data set lends itself very well to optimization through machine learnings. 200

Collaborative Filtering

Collaborative filtering is a recommendation algorithm that compares users of a product to other users and recommends new products based off likes from similar users. The idea is that if two people agree on their evaluation of certain items, they are likely to agree again in the future. If a platform can determine that User A likes Products X and Y, and User B likes…