Photo taken from "How the Spotify Algorithm Works in 2024" by Grizzly Beats LLC
Spotify has over 615 million users and over 100 million tracks that have been released by over 11 million artists. So how does Spotify know what YOU want to listen to at any given time? Spotify uses a high-tech algorithm to figure out just that. The algorithm controls every aspect of the user experience, from making personalized playlists, choosing which songs to shuffle to, giving concert recommendations in your area, what albums and playlists show up on your homepage, and much more. Spotify, first breaking onto the music scene in 2008, has changed how the general public consumes music, so let’s dive into how they do it.
Spotify uses complex AI systems to generate their playlists and recommendations. However, they have mastered the craft of making their AI more human-like by using multiple programs and computational techniques to advance that AI. Spotify bought The Echo Nest, a music analytic firm, in 2014, that combined machine learning and language processing to create a database for their music. One aspect of their algorithm is collaborative filtering. As described by Ziad Sultan, the Vice President of Personalization at Spotify, collaborative filtering analyzes trends in music listening to try and understand why songs are played together often. He describes it as a 3 dimensional map, and each green dot seen in the image below represents a different song. The position of each dot in conjunction with each other is determined by collaborative filtering. The dots that are placed close to each other are songs that are commonly played together, while dots that are farther apart are not. However, basing recommendations on collaborative filtering is not perfect. A great example of the flaws in using collaborative filtering alone is holiday music. On holiday playlists, songs like “Christmas Tree Farm” by Taylor Swift and “A Holly Jolly Christmas” by Burl Ives are commonly played together, despite one being a pop song and the other a classic Christmas carol. So if Spotify were to only use collaborative filtering, the algorithm might recommend other Burl Ives songs to Taylor Swift listeners, despite the fact that those listeners aren’t interested in Christmas carols. To prevent this from happening, Spotify uses content-based filtering.
Screenshot taken from The Wall Street Journal's video titled "How Spotify’s AI-Driven Recommendations Work | WSJ Tech Behind"
Content-based filtering files metadata (in simple terms, data about data) from each track, which includes release date, track title, artist name and any featured artists, songwriting credits, producer credits, genre, and much more. It also performs an audio analysis, in which metrics such as danceability, energy, and loudness are assessed to describe the track sonically. In addition, another algorithm in the audio analysis that analyzes tempo and splits the track into sections of varying detail. There is also an algorithm to study the lyrics, which takes note of the language the lyrics are sung in and the word choice in the song.
Photo taken from "Behind the Scenes: How Spotify Curates the Top 50 Global Playlist" by Medium
For newer and smaller artists, with the algorithm alone, it would be difficult to succeed on the platform because of the lack of user data. Spotify recognizes this, and to mitigate this “cold start problem”, they use human editors to create recommendations. They make editorial playlists such as Today’s Top Hits, New Music Friday, Rap Caviar, and much more. Artists can pitch songs on Spotify’s website to be featured on one of the hundreds of playlists editors create.
As AI and machine learning becomes more complex and optimized, Spotify is adapting these new technologies into their app. They introduced Spotify DJ in February 2023, a generative AI dj that replicates the structure of live radio. They use a human voice, named X, who gives brief messages on the reason for recommendation. They are also looking to implement reinforcement learning, which allows the algorithm to receive feedback on recommendations immediately that will diversify recommendations.
Although Spotify has released a lot of information on how their algorithm works, a lot is kept hidden. They are continuously trying to advance and optimize their algorithm to keep up with their competition, and it shows. I hope this gives you a bit of insight into the complex world of Spotify’s algorithm. Do you want to hear about the inner workings of other music streaming services? Feel free to reach out to us on our socials linked below!
Written By Lauren DiGiovanni
*copyright not intended. Fair use act, section 107.