When you’re searching for new music, you probably want to be able to find something that’s not only new, but also relevant to you. Spotifys Discover Weekly service provides that opportunity. The service utilizes machine learning to learn what music you like and find new music that you might not have heard before.
Origins of Discover Weekly
Discover Weekly is a feature of Spotify that makes recommendations based on the music you’ve been listening to. It’s like having your favorite musician in your pocket. The songs you’ll see in the recommendations are based on your personal listening habits and your friends’ playlists. This means you could hear a song you’ve never heard before.
The most popular feature of Spotify’s paid subscribers is the Discover Weekly. Each week, the service generates a new curated playlist for its members. They’re meant to make you discover new music and help artists build their careers.
According to the company, this two-hour playlist is worth more than 2.3 billion hours of listening time. So far, 40 million users have checked out the feature. A large portion of them have been U.S.-based, a fact that makes the company’s product more popular than its competition.
Another good thing about the Discover Weekly is that it gives listeners a leg up on their competition. Over 8,000 musicians had over half of their Spotify listeners come from the feature. That’s impressive, especially when you consider that many users don’t even subscribe to the service.
On Monday, the company’s algorithm updated with a curated list of new tracks. The discovery is based on the personal listening habits of the user, and a taste profile created to filter out the less exciting offerings. For the lucky few, a song may even have been on their radar months ago.
The Discover Weekly has also helped users uncover new genres, like EDM, indietronica, and indie folk. Some people have stumbled upon songs they didn’t know existed years before. These discoveries have helped artists build their careers.
In the beginning, the company experimented with various lengths and formats for the Discover Weekly, and settled on a playlist of at least two hours in length. And, it’s no secret that the company is losing money. But they still haven’t tapped into the revenue potential of this feature.
There’s one major drawback: The Discover Weekly isn’t available to mobile users yet. Hopefully, the company will roll out the feature to all devices soon.
Discovery algorithm
Spotify’s machine learning discovery algorithm for new music is an art that combines audio, Natural Language Processing, “ambience factors”, and historical user behaviour data. Compared to seven years ago, the algorithms used by Spotify are a lot more complex. In addition to analyzing the sound structures of songs, they also scan billions of playlists created by users.
The NLP model uses terms and weights to create a vector representation of a song. They use blogs, news articles, and other text around the internet to source their data. For instance, when a user mentions a song, they are likely to mention the artist and the title of the song.
Spotify’s matrix contains 30 million songs and 140 million users. Each user has different pages and playlists. This is where their machine learning recommendation algorithm comes in. It’s designed to figure out what kinds of content will interest each user.
Interestingly, Spotify does not recommend a track to every user. Instead, they recommend songs to users based on what the global average person likes. And it only rewards the machine learning algorithm that comes up with the right guess.
The best quality data is from people who love a song. Using the 30 second rule, Spotify can decide if a song will get an enthusiastic listen. If a listener does not stop playing the song after the first thirty seconds, it will classify as a good recommendation.
The’seen a lot of’ metric is still important, though. Spotify analyzes how songs are related by scanning billions of playlists. They then identify descriptive words and phrases, and classify them into cultural vectors.
Using the mel scale, they try to replicate a human hearing response. Similarly, they use spectograms to plot time-frequency patterns in an audio track.
As far as the best solution for recommending music to users goes, Spotify has taken an ode to the mascot of the recommendation industry. Their “BaRT” system, or Bandits for Recommendations as Treatments, is a machine learning-powered algorithm.
Spotify uses Natural Language Processing, or NLP, to help them discover what people are saying about certain artists. In addition to blogs, they use news articles, track metadata, and other text found on the web to source their data.
Spotifys discover BART algorithm
One of the most important aspects of music streaming services is their recommendation algorithms. Several of the top companies offer this service. Spotify has a machine learning model that uses user behaviour data and historical user data. It also keeps trust and responsibility in mind. The algorithm generates a release radar and a daily mix.
Its algorithm called BaRT, and it recommends new songs. It does this by analyzing lyrical content, music history, and the song’s features. For instance, it considers the lyrics, the music’s mood, and its inclusion in other playlists.
Using the BaRT model, Spotify can make personalized recommendations to each of its users. The algorithm analyzes the user’s listening habits, and tracks the lyrics of the songs they’ve listened to. It then compares the new tracks to the listener’s past habits. If a track is played for more than 30 seconds, it’s considered positive. However, if it’s played for less than this amount of time, it’s considered negative.
Another important feature of the BaRT algorithm is its ability to track social media and blog posts. For example, if a music blogger mentions a certain song on Twitter, the BaRT algorithm will investigate it. This allows Spotify to find similar bands or tracks.
While the BaRT algorithm makes personalized recommendations, it can also make mistakes. Some examples of this are when it recommends a song after a listener has skipped the first thirty seconds. Keeping the skip rate low is important, and it can achieve by grabbing the listener immediately.
Spotify’s algorithm is constantly evolving, and it’s always working to provide its users with the best experience possible. With that in mind, Spotify offers its users an algorithmic playlist called Discover Weekly.
During its initial stages, the BaRT model used all the information it had on the listener to make a recommendation. This done by considering the individual’s previous listening habits, their click-through rates, and their satisfaction levels. But over time, the algorithm started to take into consideration their actual preferences.
The exploitation and exploration models are two of the different systems that influence the way BaRT recommends new songs. These two modes are helpful when it comes to recommending content to new users.
Spotifys discover personalized experience
Spotify’s Discover Weekly service is a music recommendation tool that generates a curated playlist of new tracks based on your music tastes. This personalized playlist sent to users every Monday. Users can like the songs they receive, add them to their library, or save them to a playlist.
To personalize the experience, Spotify utilizes machine learning. Its audio models analyze raw audio tracks, identifying patterns between tracks. The songs then grouped together by shared attributes. During the process, Spotify also uses its NLP model, which constantly trawls the web for articles, blog posts, and other text about music.
Machine learning algorithms use data aggregation methods to recommend tracks based on your preferences. The music platform also uses NLP to build a profile for each song. A neural network can predict the tempo, acoustical characteristics, and even the personality of the song.
Moreover, a virtuous circle of data generated through user engagement. As a result, larger data sets lead to improved recommendations.
In addition to recommending songs, the service also provides a feature called Enhance. This helps users retrieve recommendations within their own playlists.
Discover Weekly playlists are based on a combination of listening trends, user preferences, and other criteria. For example, if you’re a big fan of hip hop, the service will send you a playlist of music containing songs from popular rappers.
Similarly, the Discover Weekly feature can also recommend tracks that are like your own favorites. Using machine learning, Spotify’s ML algorithms can analyze user behavior, speakers’ preferences, and genres. They can also identify parallels between user mood and the songs.
The Discover Weekly curated playlist sent on a weekly basis, and will contain 30 new songs. Users encouraged to listen to the playlist, because it will give them a taste of what’s new. However, it’s important to note that users given a maximum of two hours to listen to the songs, which is plenty of time to enjoy an immersive listening experience.
Unlike Apple Music and Google Play, Spotify’s Discover Weekly feature is not based on recommendations from third party sources. Instead, it built on three main models.
