Music ex Machina: Do Recommender Systems Homogenize Music? (Job Market Paper)

Recommendation systems are the driving force behind many of the suggested products we see online, from retail purchases to video and music streaming. While a lot of research examines how these algorithms impact pricing and consumption, not much looks at how they influence the actual product design and characteristics.

In my job market paper, I investigate whether music recommendation engines like Spotify’s encourage more homogeneous, similar-sounding songs as artists try to cater to the algorithm’s preferences rather than diverse listener tastes.

Using a structural model analyzing interactions between consumers, the recommendation system, and forward-thinking music producers, I leverage data from Spotify to estimate the effects of their recommendation engine on song traits like length, danceability, and acousticness.

Some early findings: Consumers gravitate towards newer, more danceable tracks, with premium subscribers less likely to skip songs. Interestingly, song length doesn’t seem to significantly sway listeners. However, Spotify does appear to promote shorter, more acoustic songs through its recommendations.

This hints that artists may feel compelled to make their music briefer and more mellow to please the recommendation algorithm, even if consumers don’t particularly favor those attributes. My ongoing supply-side analysis aims to quantify this potential for reduced variety in music creation.