The Algorithm-Based Recommender Systems For Music & What Artists And Labels Should Know

By Jan Mohr

The importance of algorithms should have been something that came across nearly every active content creator or consumer in their online environments.

An algorithm is a defined set of repetitive tasks to collect information and do a computation in order to solve a given problem. There are plenty of examples out there from search engine results, encryption for data protection over to machine learning for AI and of course user recommendations on platforms like TikTok, Instagram, Facebook, Netflix or YouTube and Spotify.

YouTube and Spotify are probably the most popular and most discussed examples when it’s about music recommender systems, but also on Apple Music, Amazon Music, Deezer and others are algorithms working – even though Spotify and YouTube seem to apply the most advanced technology these days.

When having a quick browse through quotes from researches and platform statistics you can find numbers like this:

  • 62% of people say streaming services are among their top music discovery sources (according to the new Music 360/2020 report by MRC Data/Nielsen).
  • 1/2 of these discoveries approximately come from recommender systems.
  • 1/3 of artist discoveries on Spotify by “Made for You” recommendations (according to the recently released Made to be Found report).
  • 70% of content watched on YouTube comes from recommender system (according to YouTube product officer in 2018).

The role and importance of music streaming and their algorithmic recommender systems seem to become obvious.

When looking at the Digital Service Providers (*DSPs) for music out there, the algorithmic recommendations have to understand the data they recommend and the user they recommend it to.

While Spotify applies a quite complex scan to describe audio files, YouTube seem to rely on the given descriptions from the content creator and user behavior to classify videos. What all have in common is to build and compare user profiles, where playtime, follows, adds, likes, shares etc. get considered. Furthermore, the metadata from the distributor’s delivery and information from pitches have an impact on recommender systems.

Big pools of user and content data allow quite complex comparisons these days, which already became sophisticated scientific data research. A rule to always keep in mind is: Platform algorithms get adjusted according to platform goals.

When looking at the development of the YouTube algorithm it all began with the goal to attract views and clicks. This moved over to increase watch time and platform stay, while trying nowadays to increase user satisfaction.

This means the algorithms became more personal. Also, on Spotify it is about to recommend the right music for the right person at the right time.

The overall goal is: user satisfaction and engagement.

A recent discovery of algorithmic DSP improvement endeavors is also that people tend to be most satisfied when their long-term listening experience is diverse. This means they still want to find similar music to what they already like, but also want to discover new and fresh music over time.

Ok … and what are the findings for today’s labels and artists?

Use of organic and paid traffic opportunities can help to build a solid performance.

Each user, who adds music to a playlist or follows an artist is a clear sign for the recommender system, no matter where the user came from. This includes all external sources that drive audience to artist profiles, such as ads on social media, media coverage, influencer shares and so on.

But please be aware of dubious offers that guarantee playlist inclusions and to increase streams in exchange for money. There are service providers, who use bots (fake profiles). These are violating against streaming platform rules. If such activities get detected on an artist’s profile it can lead to content removal and in the worst case a ban of the artist from the platform.

The algorithms learn more about the content during its life on the platform.

This means a recommender match tends to be improved over time. To keep on highlighting and promoting content after its release date therefore still helps to train the algorithm and to get music delivered as recommendation.

The algorithms don’t care too much about the release date.

Even if new music is interesting for a diverse and thus satisfying listening experience of users, not the actual release date decides. A song that is not new to a platform can still be new to a user.

The recommender system prefers quality content.

Finding your niche and having well-done output is more important than uploading with a tight frequency. Even if regularity to a release schedule can be helpful to maintain user engagement on artist profiles, it seems not the most important influence for the algorithms. When having a look at user recommendation on Spotify via “Discover Weekly”, “Release Radar” or “Your Daily Mix”, not how new a song is but how enticing it appears to a user seems to be most relevant.

Build up a catalogue of quality content can help the algorithmic recommendations.

The bigger a release catalogue is, the more information it potentially provides for the algorithms to learn when to recommend. Since the recommender systems will always share the best match a solid data source can make the match more reliable to the recommender system. (Btw. there are plenty of established labels and artists, who receive their biggest share of revenue from their back catalogue proceeds).

Also new artists have a chance to get recommended.

The platforms want to have their users satisfied, which on a log-term perspective means to also offer them new and fresh discoveries. As a result, being a new artist can potentially make the right match, if the algorithm finds the right information.

(To help your profile and music become recommendable, please also pay attention to the next points).

Be accurate and concise with your metadata and descriptions.

Every recommender system takes the delivered metadata into account. To make it accurate helps to generate the right match. 

The pitches via Spotify for Artists help their recommender system.

The pitches sent via Spotify for Artists provide additional information such as mood, style, cultural background, instrumentation, song description, etc. which directly help the algorithm to categorize and recommend the music. This can be helpful for new artists to improve possible cold-start problems on the content recommendation.

Last but not least: Don’t forget about the editorial teams!

There are still real people listening to music and building handmade recommendations and playlists. These editorial playlists have a huge followership and as a direct result also a strong impact on listening behavior of platform users. This means they also influence how the algorithmic recommender systems evaluate music.

To be considered for these types of playlists is always an editorial decision from the DSPs.

Since this is still a strong influencing factor on performance the Paradise Worldwide team has built and constantly maintains a solid network to editorial staff of the major platforms to directly pitch and highlight new music. (So, if you are a client of ours, please always stick to our deadlines for marketing drivers and try to consider our tips for improvements. 🙂