How are music artists adjusting to algorithmic recommendation systems?

I have written a few times, here, on the blog, about algorithmic recommendation systems. Sometimes, I focus on the technology itself. Other times, I focus on the recipients of those recommendations. Occasionally, I have looked at the impact on advertisers. However, I don’t often consider the perspective of the producer – i.e., the person / company / brand that creates the content who distribution is mediated by the algorithm – for instance, films, music or news.

Thus, I was looking forward to reading the report “The impact of recommendation algorithms on the UK’s music industry”. It was authored by researchers at the Centre for Data Ethics and Innovation, for the Department for Digital, Culture, Media & Sport. The report includes the findings from a small-scale survey with music creators. The sample was quite small (102 respondents) and not representative. So, the findings need to be treated with much caution. The research team also conducted some interviews, though, unfortunately, there is no detail in the report about the method (e.g., how many interviews), and very little about the findings.

As I discussed in this blog post, internet technology reduces search and transaction costs for music buyers, as well as the costs of production and distribution for musicians. This combined effect was expected to increase diversity in the music industry. However, US musical top charts data (at the time of that blog post) showed a different picture: rather than increasing diversity, digital technologies seem to have created bigger superstars (Figure 1).

The creators surveyed in the Centre for Data Ethics and Innovation’s study expressed a similar sense of disappointment with the potential of music streaming services to enable them to reach wider audiences. As shown in figure 2, below, only a third of the artists felt that the recommendations on music streaming platforms had enabled them to do so. The interview data seems to indicate that the sentiment is spread across small and large labels (though, again, we lack detail about the interview responses to be able to understand who is benefitting, how or why).

More than half of the sample reported that they had changed how they released their work, in order to increase their likelihood of being recommended by the system (Figure 3), for instance, by careful selection of tags to classify the track. In a way, it isn’t much different from other distribution channels, with “tags” substituting for shelves, or genres, or sections in store.

On the other hand, most content creators said that they had not changed how they made music, to “play” the algorithm (Figure 4). According to the report, this is due to “a lack of knowledge about how recommendations are made on streaming services”.  Though, the report also highlights the “anecdotal evidence about creators more frequently collaborating with one another, in the belief that multiple artists being tagged on a track would make them more likely to be recommended and to a wider audience” to suggest that, indeed, there have been some changes in the type of music produced, lately.

The key to succeeding in the age of algorithmic recommendation systems is, according to the report, experimentation: in the absence of transparency from the streaming companies about how recommendations are made or how consumers are engaging with their music, creators need to experiment with different techniques to try and get their work picked up by the algorithms. Those able to “experiment with different ways to optimise their music for recommendation” would be at an advantage. And, in the absence of data, this means that “creators with greater resources would be better placed to experiment” and, thus, succeed.

In my view, the report raises more questions than answers. However, it is still an interesting read. It includes a short overview of literature on the topic, as well as a poll of 4,000 UK music consumers. 

Do you feel that you discovered a new artist (i.e., found it interesting and went on to explore their work further) because of an algorithmic recommendation system?