Who Is Not Recommended On Spotify?

We are analysing how recommendations work for the artists involved in our Listen Local experiment. We created a testing database with their music and other artists from their cities, regions and countries. We need new recommendation engines to dig out from a pile of international hits the local artists.

  • Avoid landing on Forgetify, in the universe of never listened-to-songs.
  • Understand how artists and their songs can find audience in the region relevant to their career goals, tour destinations, export opportunities.

Our preliminary analysis is not representative for nationality, though we believe that we will find serious regional, linguistic and national patterns. We are using examples from Spotify. To be fair to Spotify, while their algorithm has many problems, their API is at least very open, as opposed to Apple Music where we do not really know what is going on.

Our main concern is that some artists and their songs seem to be never recommended. Moreover, if artists and songs from a city, region or country are rarely or never recommended, then they risk losing their home market: the local audience will be recommended music that is coming outside of the city, region or country.

One of the most important way recommendations work is via related artists. If an artist has no related artists, it is very unlikely that either via user browsing interaction or via the algorithm that artists or her songs will be discovered by anybody.

If you want your music and audience to be analysed in Listen Local, fill out this form. We will include you in our demo application and our analysis to be revealed in December.

Because big data algorithms learn from user behaviour, more popular, more often selected songs, and songs selected by more users are more likely to be recommended.

If the algorithm finds more related music, it means that it “learned” the repertoire of the artist. Artists with very little popularity tend to be related to 20-30 artists, but at a popularity level of 40-60 they are related to 30-40 artists. Seems simple, right?

Not that much. If you want the algorithm to learn your music well, you really have to go “viral”.

Doubling your routes to being recommended via other artists' music requires exponential growth. You need to have 1000x more followers to double the routes leading to your artist profile and recordings.

As we have shown, the current recommendation engines are re-enforcing the existing status quo, and they can re-enforce the marginal status of subcultures, artists from a nation, city or region. In other words, as we often say, big data creates injustice.

With our Music Observatory for larger organizations and with our Listen Local analytics and recommendations for artists, managers and labels, we want to fight exactly this injustice. We are creating transparent, open source, and responsible tools and algorithms.

Related