/
6 mins read

How Spotify Suggests Songs You Like

person using smartphone
Photo by John Tekeridis on Pexels.com

In the age of the Internet, the most successful websites have managed to be successful because they offer people what they do. Need. Or at least, they make them believe they need it. The more time a person spends on a website, the more likely they are to spend money on it. A clear example is Spotify, but we can say that it all started with Youtube. The video platform needs us to see the ads of the videos and in that way more benefit. The longer the more money

In this article I’m going to talk about how Spotify has a similar approach, but instead of offering us one video after another, it offers us one song after another. Spotify does everything possible so that we listen to more and more music and be connected to its platform for a long time.

Spotify recommendation and search

Spotify’s mission is to promote music and artists. They offer a platform where creators offer their musical works to be enjoyed by their fans or anyone who wants to listen to their music. That mix of fans and artists is the heart of Spotify, since the platform will try to offer you music relevant to you based on your tastes.

In the same way that when you surf the net and find ads relevant to you since web pages track what you do, will also try to reward you by offering what best suits your musical tastes. It starts from an immense catalog of songs, then there is a series of playlists (list of songs), podcasts, etc., that classify the songs in the catalog. On the other hand, the user searches for the songs he likes and it is that choice that activates Spotify suggestions.

Information retrieval and information filtering are, in fact, two sides of the same coin  (Nicholas J. Belkin and W. Bruce Croft, 1992). Spotify has this claim in mind. Since the recommendation and search system works, data recovery and filtering are low. They work together to deliver music that the listener enjoys.

Spotify home screen, the push paradigm

The Spotify app’s home screen is a prime example of how algorithms govern an auditory experience. Its goal is to quickly help users find something they like to hear and hear right away.

The home screen is governed by a machine learning algorithm called BaRT (“Bandits for Recommendations as Treatments“). The task of the system is to organize each home screen in a personalized way for each user. That includes the “shelves,” or rows of playlists, that follow a theme like “the best of artists” or “keep the vibe,” and the order in which playlists appear on those shelves. It therefore helps to listen to music right away.

The Bart algorithm allows you to jointly customize recommendations and associated explanations to provide more transparent and understandable suggestions to users. Bart is an effective method to address the problem of exploration-exploitation in recommendation.

Spotify uses machine learning algorithms when the user is not connected, offline, which make decisions for when they are back online and offer them the right playlists.

Search, the pull paradigm

Having a huge catalog of songs makes it necessary to have a suitable music search mechanism. Spotify owns more than 80 million songs in 79 different markets. There are different types of modalities such as podcasts, music, audios, etc. Finding what the user is looking for is the heart of the pull paradigm.

When you search it instantly offers you the best suggestions letter by letter taking into account the tastes of the user. Listen and organize the information to offer you the most suitable for you.

User data September 2021

If you are interested in numbers, this section is precisely about them. In the third quarter of 2021  all the main indicators of Spotify’s business have ended very satisfactorily for the company.

Monthly active users grew 19% year-on-year to 381 million  in the quarter, up from  365 million last quarter. Spotify had double-digit year-on-year growth in all regions, with particular strength in the rest of the world (the world except Europe and the Americas), where performance was supported by the resumption of marketing activity in India along with above-plan growth in the Philippines and Indonesia.

Premium subscribers grew 19% year-on-year to 172 million in the quarter,up from  165 million last quarter.

References

  • https://onezero.medium.com/how-spotifys-algorithm-knows-exactly-what-you-want-to-listen-to-4b6991462c5c
  • https://www.slideshare.net/mounialalmas/recommending-and-searching-spotify
  • https://ciir-publications.cs.umass.edu/pub/web/getpdf.php?id=131
  • https://research.atspotify.com/publications/explore-exploit-explain-personalizing-explainable-recommendations-with-bandits/
  • https://static1.squarespace.com/static/5ae0d0b48ab7227d232c2bea/t/5ba849e3c83025fa56814f45/1537755637453/BartRecSys.pdf
  • https://investors.spotify.com/financials/press-release-details/2019/Spotify-Technology-SA-Announces-Financial-Results-for-First-Quarter-2019/default.aspx
  • http://dx.doi.org/10.1145/138859.138861
  • http://pchandar.github.io/static/Gruson2019-a2c9a8576182dcb33019d20a7c7a51b7.pdf
  • https://twitter.com/thexxlman/status/1455140141970903043?s=21

1 Comment

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Previous Story

Crear obras de arte con inteligencia artificial

Next Story

Cómo Spotify sugiere canciones que te gustan

%d bloggers like this: