How Spotify Knows What You Want to Hear

The BaRT Algorithm Explained

Did you ever wonder how Spotify always knows what music to play next? And what do you think of the Daily Mixes? Six personalized lists sorted on genres. The only thing you have to do is pick a list that matches your mood and you’re ready to go. How does Spotify create these lists? How does it know what music you will like?

Introducing BaRT

Meet BaRT, the Spotify algorithm. The brain behind the music suggestions. BaRT stands for ‘Recommendations as Treatments’. BaRT’s goal is to help you find music you will probably like. And the better it is at his job, the more likely you will keep using Spotify. To make this personalized experience possible, BaRT uses a lot of data. Data from music, data from you, and data from other customers.

What Can BaRT Do?

Three systems influence BaRT’s music recommendations: Audio Analysis, Collaborative Filtering, and Natural Language Processing. How these systems work together or which one is more important is a blackbox.

1. Audio Analysis

The most straightforward part of the Spotify algorithm is the Audio Analysis. BaRT looks at the different musical characteristics of the songs you listen to. For example, what is the tempo, which instruments are used, and how danceable is the song? By comparing these metrics, BaRT can find a pattern and recommend similar music. This way, it can also create specific playlists with only upbeat songs or relaxing songs. This is ideal if, like me, you have a diverse music taste, but like to listen to certain music genres depending on your mood.

2. Collaborative Filtering

The second system is called ‘Collaborative Filtering’. The algorithm analyses the behavior of other users who like similar music. What kind of songs do other people like? Are there new songs you don’t know yet? By looking at what other people like, BaRT tries to find music you will probably like. So basically, Spotify users are indirectly help each other out in their search for good music.

3. Natural Language Processing

The BaRT algorithm ‘understands’ language through Natural Language Processing. It can analyze the lyrics and the language the song is written in. And just like Google, BaRT can crawl the web. With its ability to process language, it can read what people are saying about music on the internet. BaRT looks at blog posts, news articles, and social media to see what people are writing about certain artists or songs. Do some artists work together? Did a certain song get a lot of publicity? BaRT uses this information to find new similarities and to promote popular music.

Do you Have Influence?

In addition to the three factors mentioned above, you can also influence what BaRT will recommend. Your behavior has a big impact. Which songs do you skip? Which songs do you save? How often do you repeat a particular song? But also: which artists do you follow? And it goes even further. Spotify keeps track of how long you look at, for example, an artist page or an album. After processing this information, BaRT quickly analyzes your behavior and the behavior of people with a similar taste in music and then recommends a song that you can’t say no to.

Final Thoughts

You may be surprised because of the different kinds of systems and data processing methods of the Spotify algorithm. However, BaRT does what it’s supposed to do: offer you a unique music experience. Personally, I love that you get new music suggestions and you can choose a music list based on mood. And when you don’t like a song, you can simply skip it or switch to your own Spotify lists. The great thing is that BaRT learns when you change your behavior. If you choose other songs, BaRT will notice and will probably integrate it into your new Daily Mixes.