Here’s how Spotify’s algorithm recommends new music

If you use Spotify, then chances are that you have opinions on its recommendation algorithm. I quite like it, but I’ve noticed a significant amount of repeated recommendations, particularly of music that I have listened to a lot of already. As it turns out, Spotify has actually outlined how its recommendation algorithm works, which could potentially help you to know, too.

How Spotify’s recommendation algorithm works

There’s a lot of machine learning

Spotify has outlined the basics of how its recommendation algorithm works, and these are the technical details to break down that simplified information into what it actually means. These are the three pillars of recommendation that Spotify uses:

• Collaborative filtering: This model suggests content based on similarities between users and their behavior. If two users listen to similar songs, Spotify may recommend additional tracks one user likes to the other. Collaborative filtering excels when data on multiple users with shared tastes is available.
• Content-based filtering: This model recommends songs based on the characteristics of the content itself. Spotify categorizes tracks based on metadata (e.g., genre, tempo, mood) and may use Natural Language Processing (NLP) to analyze lyrics or external descriptions. If a user often listens to upbeat pop songs, content-based filtering can suggest similar-sounding pop tracks.
• Deep learning models/Neural networks: Spotify uses neural networks to identify complex patterns in user data and content features for personalized playlists like Discover Weekly or Release Radar. Recurrent Neural Networks (RNNs) are often used in sequence-based recommendations to predict what a user might want to hear next based on their listening history.

The first is a very common strategy employed by other streaming services too, most notably Netflix. To give an example, if one user enjoys the same shows as another user, Netflix will recommend the same shows the other person likes to the initial person. On top of that, Spotify also looks for specific characteristics, which we know from the company’s yearly Unwrapped event that it tracks a lot of data about the music shared on the service. Using Spotify’s API, you can even look for parameters like “danceability.”

Spotify also uses similar techniques when it comes to search. Spotify says that if many users searching for something interact with a specific result, this will then inform the results of future searches for future users. The company’s neural networks also build your “taste profile,” and actions like searching, listening, skipping, or saving to your library inform Spotify what you like and don’t like.

Finally, Discovery Mode allows artists and labels to highlight specific songs for increased visibility, basically prioritizing these tracks in the recommendation algorithm. However, this isn’t a guarantee of placement; Spotify still ensures that any promoted content aligns with your taste profile and engagement patterns. If a Discovery Mode song doesn’t resonate with listeners, Spotify’s algorithm will naturally deprioritize it. This works the same way as regular targeted advertising does, where it will only show content to you if it matches your interests.

Related

How I built a Spotify alternative with a Raspberry Pi and Navidrome

Not a fan of Spotify’s subscription model? You can use Navidrome to turn your Raspberry Pi into a music-streaming server

How can you influence Spotify’s algorithm?

Spotify gives you a few tools to do so

If you want to guide your recommendations, you can give Spotify signals about your preferences:

• Exclude from your Taste Profile: You can exclude specific playlists from influencing your taste profile, helping Spotify understand that these particular listening habits don’t necessarily reflect your overall music taste.
• Mark songs as “Not Interested”: Actively marking a song or artist as “not interested” reduces the likelihood of similar content appearing in your recommendations, giving you more control over what you see.
• Explicit content filter: If you prefer to avoid explicit content, Spotify’s explicit filter will block it, excluding it from your recommendations entirely.

Basically, Spotify’s recommendation algorithms work in the same way that you would expect from any recommendation algorithm. Engage with the tools the service provides you, and you can change how it recommends tracks to you.

Spotify’s algorithm is more than just a playlist generator, it’s a music discovery service that can help you to find new artists and songs that resonate with your listening profile. Not everything will suit you of course, but I know that it’s helped me find music and discover new artists that I never would have otherwise.