How does a music recommendations algorithm work to boost user engagement?

Do you use an online music-streaming service? Probably a silly question given the popularity of these apps, but how else would we start a post like this?

So good; you’re familiar with “intelligent” music recommendations for songs and artists. Maybe Spotify, Apple Music, Noteflight, Amazon Music, Pandora, or another music suggestion tool has helped you discover a new favorite singer or new album. Maybe the melding of machine learning with music has made you throw out your CD collection because it can’t match the online experience, which lets you listen to various artists at the same time, among other options.

Goodbye brick-and mortar shops, hello music apps 

Before streaming platforms came along, discovering new music meant music lovers traipsing into a music store and flipping through the various categories of items. Now, as is the case with many other consumer products, song recommendations and the platforms that promote them have fully revolutionized the world of music.

Music fans can delight in song discovery “for me” right on their mobile phones, without leaving home, or while doing daily activities, as they savor the sounds of their amazingly well personalized recommendations.

Modern digital music recommendations do what your best friend might have done in the past (you know, like the one in high school who’d bring in a new CD to sample every week). They say “Hey, listen to this. We know your tastes and feel that you’d be into it.”

And usually, they’re right. That’s because music recommendation functionality is (pretty obviously) powered by music recommendation algorithms. 

The algorithms make it easy. They can throw up new gems to be discovered and bring listeners back to their old favorites from decades past. They can keep users engaged and continuing to use the app. They can help artists find new audiences while  connecting listeners with lesser-known tracks and emerging albums.

The mystery of how a music algorithm works its magic or how your song recommendations are linked — that is, related to each other — is much less obvious, if impossible, to figure out. It’s data science; that much is clear.

How a music algorithm works 

Music recommendation algorithms are just like any other recommendation algorithm. They’re designed with mathematical formulas that are usually AI driven.

As with any other item recommendation tool, the purpose of music-recommendation algorithms is to create a great user experience. The goal isn’t to generate sales or convert users to a particular music genre (although this may happen as a side effect). The algorithm instead helps create an enjoyable personalized experience to help keep people engaged in the act of listening, thereby generating the revenue supplied by their continuing to subscribe.

Unlike other types of search engine, music recommendation software makes suggestions based on artist and genre similarity. It looks at preferences (e.g., frequently listened to songs in the user’s library and their favorite artists) and also takes into account dislikes/aversion (skipped songs, ignored genres) to build a comprehensive profile. Music platforms that operate based on music purchases (e.g., iTunes) rather than subscriptions also take into account users’ buying history in addition to their listening history..

Spotify’s winning approach

The Spotify recommendation algorithm has received accolades. It examines song structure and figures out how music is related by scanning its user-created playlists. In addition, it hypothesizes about members’ musical preferences by analyzing their listening habits.

The result: new songs appear at the bottom of your playlist, and they seem to be connected based on an artist or genre. Other times, you’re given a  whole new Spotify playlist. There seems to be a connection to your listening preferences, but you can’t figure out what it is. 

The Spotify music recommendation functionality also relies on whether you listen to the recommended songs on your home screen.

The algorithm looks at the length of time you spend listening, and if it’s more than 30 seconds into a song, that’s noted as a thumbs up. And the longer you spend playing a song or playlist, the more tailored to that user data your subsequent suggestions become. If Spotify’s robot occasionally misses the mark on your recommended songs, your every move is still tracked by the music recommendation software and noted so that it can keep steadfastly aiming to please you so you’ll keep your Spotify account.

Recommendations: let us count the ways

With music-streaming services, suggestions are provided to listeners in a variety of ways, even if they’re all doing basically the same thing: making recommendations.

There are two types of recommendations: personalized and non-personalized. Many people would argue that with music, as with anything these days, personalization is always the way to go, but some streaming services may also see a benefit in recommending music more generally.

Here are some of the recommendation methods used by the various music services: 

Personalized recommendations

  • Discovery: It’s fun to unearth music that moves you, isn’t it? Whether it’s a track, album, or artist, music discovery functionality helps users stumble upon new tunes that match their preferences. For example, someone who’s indicated a love of country music will be shown recommendations to other good music in that genre, perhaps including popular songs from an up-and-coming artist. Spotify’s Discover Weekly is one example of this feature. 
  • Enjoyment prediction: This more “assertive,” crystal-ball music recommendation engine technique utilizes AI to assess whether a user will enjoy a particular song, artist, or album. The result? You may be amazed by the service’s ability to read and delight you.
  • Playlist generation: A cultivated playlist based on the listener’s similar tastes. If you like various kinds of music to put you in different moods, you’ll get playlists geared for that.
  • Top content: A selection of the listener’s favorite songs. If you don’t want to have to change the channel because you hate the sound of something, this is your type of recommendation. Apple Music Replay is an example.
  • Radio streams: The word “radio” can throw you, as on Spotify, at least, you don’t get actual radio stations (like with call letters and interruptions for ads). Instead, this is simply another way of presenting playlists based on preferences. If you’re listening to a song you like and you click “Go to radio,” you get a bunch of similar songs, which you can sample and add to your own themed playlists. Or just keep playing your radio station.

Non-personalized recommendations

  • Trending list: A list of popular items (e.g., top artists, songs, soundtracks, albums).
  • Curated playlist: A song list that’s not relevant to a user’s preferences but could still lead to their discovering of new songs, artists, and genres.
  • Radio streams: Recommended sets of random songs that aren’t relevant to a user’s preferences, either. 

What are the benefits of music recommendation software? 

A better question would be what aren’t the benefits, and who doesn’t benefit from receiving AI-generated song suggestions?

Music and song recommendations improve the usability of any music streaming platform, connecting music to the people who want to hear it and thereby boosting user engagement. And that’s key, because while an ecommerce store has in its catalog thousands of items at most, a music-streaming platform may have access to millions of songs (with some songs streamed billions of times). Without a music suggestion tool, finding new music that the listener could enjoy could be a pretty difficult task.

Here are the key benefits:

Personalization for listeners

Personalized content hits home with consumers; there’s no question about it. Curated content that’s “just for you” is sure to create a feeling of warmth while using the app. And then there’s the fun of  listening to your own playlists that you know will put you in a good mood, whether they’re Spotify’s “Your top mixes” or one of your personally created theme playlists (like maybe “Background working” or “Running”).

Discovery for listeners

Music recommendation algorithms generate a “curious” mentality, prompting listeners to enter down a melodic rabbit hole. For media platforms, discovery is especially important, as listeners feel like they’re broadening their horizons and cultivating diverse new musical tastes, and all of this activity can then be shared with friends. 

Engagement with content

Music recommendation algorithms create instant engagement by inviting users to sample their myriad personalized suggestions. The “thoughtful” recommendations keep people tuned in, with their ears cocked for anything that will knock their music-appreciation socks off. 

Support for musicians

Music suggestion tools don’t just benefit their listening customers and the streaming service; they benefit musicians as well, because the algorithms connect artists to a prospective new audience. Country-music singers are suggested to country-music listeners. New Pop tracks are discovered by a younger demographic. Michael Bublé finds his way to more middle-aged moms.  

Listener retention  

A stream of music recommendations that resonates with the listener creates interest and stokes their loyalty for the long term. Algorithms that really “get” a customer’s favorite types of music generate positive feelings around using the music platform, and this keeps music fans coming back. 

Business success

It’s all about positively impacting the music recommendation system’s bottom line, of course. For streaming platforms that promote music purchases, recommendations are a clear source of revenue. For subscription-based platforms like Spotify, higher engagement rates (created by users exploring and adopting new music interests) lead to the earning of more ad-based revenue.

Music to your business’s ears

Want help with your artificial intelligence–based music recommendations? With Algolia’s recommendation engine, you can do collaborative filtering, merchandise, rank, and contextualize music recommendations to reflect your unique brand and revenue goals.

Want to see how easy it is to get started with up to 10,000 records and search requests per month? Try us out by building a music recommendation API solution with our free plan, or if you’d prefer, ask us the best way for you to get started.

About the authorCatherine Dee

Catherine Dee

Search and Discovery writer

Recommended Articles

Powered by Algolia AI Recommendations

What is a product recommender (or product recommendation engine)?

What is a product recommender (or product recommendation engine)?

Catherine Dee

Catherine Dee

Search and Discovery writer
How movie and video recommendations work, and why streaming services must get it right

How movie and video recommendations work, and why streaming services must get it right

Catherine Dee

Catherine Dee

Search and Discovery writer
What are personalized recommendations and how can they boost engagement and conversion?

What are personalized recommendations and how can they boost engagement and conversion?

Catherine Dee

Catherine Dee

Search and Discovery writer