A Deep Dive Into Spotify's Discover Weekly

Do you know what the second best thing to hitting the jackpot is? It’s discovering new music you never knew was missing from your life. Every Monday morning, Spotify users are gifted with a curated playlist, consisting of 30 new songs that Spotify thinks you will love. This feature is called “Discover Weekly,” and it’s unique to each user. Let’s take a deep dive into how Spotify’s algorithm beautifully curates this playlist.

The Algorithm

Many Spotify users question how the app can accurately pick out songs that so perfectly fit their taste. Well, the answer is, there are three different types of recommendation models powered by machine learning to generate the songs that make up your Discover Weekly—collaborative filtering, natural language processing (NLP), and convolutional neural networks (CNN).

Collaborative filtering

Whether you are crying while screaming, “All Too Well” again for the 10th time this week, or listening to your favorite songs while getting ready to conquer your day, you are also helping Spotify recommend music to other users.

Collaborative filtering “Uses your behavior and that of similar users,” Clark Boyd explains. This model analyzes your favorite songs, artists, and playlists. It then takes that data and compares it to all of Spotify’s 406 million other users. Once the model finds your music soulmate, it’ll recommend you their favorite songs that you have not listened to yet.

Natural Language Processing (NLP)

Of course, Spotify doesn’t just rely on the matchmaking of other users' data. It also uses natural language processing (NLP) to create a profile for each song on its platform. Akshad Tambekar states, “NLP is the ability of an algorithm to understand speech and text in real-time.” He then proceeds to explain, “Spotify’s NLP constantly trawls the web to find articles, blog posts, or any other text about music, to come up with a profile for each song.”

The profiles created for each song consist of adjectives that are used to describe the music. Each word is then given a weight, “...reflecting its relative importance in terms of how many times an individual would attribute that term to a song or musician they like,” Ipshita Sen explains. The NLP will then analyze the profile of songs near and dear to your heart to find new songs that have identical profiles.

Convolutional Neural Networks (CNN)

Now, you may have discovered a new song on your playlist, and when you click on the artist, you’ll see they have less than 10,000 listeners a month. You can thank convolutional neural networks for that because now you have a new artist to gatekeep.

CNN is used to assure less popular songs and artists get recommended to you. This model creates raw audio for each song which then turns it into a waveform. Akshad Tambekar explains, “These waveforms are processed by the CNN and are assigned key parameters such as beats per minute, loudness, major/minor key, and so on.” New and popular songs are compared by their waveforms. The new songs appearing on your playlist will have very similar wavelengths as your previous beloved music.

I think I am finally excited for Monday morning to come around. What could be better than discovering new music that is curated specifically for you? If you’re not already an avid Spotify user than you definitely should be now.

Do you listen to your Discover Weekly playlist? Do you love the songs that are added into it? Let us know in the comments down below!