Study finds algorithmic music recommendations don’t work so well for hip-hop and hard rock

Genre of background music in restaurants can affect algorithmic music recommendations, influencing user satisfaction and engagement.

A new paper has found that algorithmic music recommendations don’t work very well for fans of genres like hip-hop and hard rock.

The study, authored by researchers from Austria and the Netherlands, appeared on March 29 in the journal EPJ Data Science.

The researchers compared the accuracy rates of algorithmic music recommendations for mainstream and non-mainstream music listeners.

To do so, they used a dataset with the listening histories of 4,148 users of the music-streaming platform

The researchers divided these users into two groups.

One group mainly listened to mainstream music, and the other group mainly listened to non-mainstream music.

Based on the artists that the users listened to most frequently, the authors created a model to predict how likely the users were to enjoy music recommended by four common music recommendation algorithms.

Music recommendations most accurate for mainstream tastes

They found that fans of mainstream music received more accurate music recommendations than fans of non-mainstream music.

The authors then created a separate model to categorize the non-mainstream music listeners based on the music they listened to most often.

They came up with the following categories: acoustic (for example folk); “high-energy” music with vocals (such as hard rock and hip-hop), acoustic music with no vocals (for example ambient), and “high-energy” music with no vocals (for example electronica).

The authors compared the listening histories of each group.

They then identified which users were most likely to listen to music outside of their favorite genres.

They found that people who mostly listened to music such as ambient were the most likely to also listen to music preferred by fans of hard rock, folk, or electronica.

People who mostly listened to high-energy (vocal) music — for example hip-hop and hard rock — were the least likely to also listen to music preferred by folk, electronica, or ambient listeners.

Nonetheless, this group did listen to the widest variety of other genres, for example punk and singer/songwriter.

Recommendation success for the “non-mainstream” listeners

The authors also tested how likely the different groups of non-mainstream music listeners were to like the music recommendations that the algorithms came up with.

They found that people who listened to mostly high-energy (vocal) music received the least accurate music recommendations.

People who mostly listened to music such as ambient received the most accurate recommendations.

“Our findings suggest that many state-of-the-art music recommendation techniques may not provide quality recommendations for non-mainstream music listeners,” said co-author Elisabeth Lex. “This could be because music recommendation algorithms are biased towards more popular music, resulting in non-mainstream music being less likely to be recommended by algorithms,” she said.

The authors suggest that their findings could inform the creation of better recommendation tools for non-mainstream listeners.

Doing so, they write, would lead to “more prominent exposure of (long-tail) music artists due to a better-connected recommendation network.”

More recent science and psychology news:

  • A new study has found that about two-thirds of people abandon their New Year’s resolutions within a month.
  • A new paper looks at recent advancements in smart contact lenses, envisioning applications well beyond fixing things like nearsightedness.
  • Researchers have shown that animal DNA — including human DNA — can be collected from the air, via “environmental DNA.”
  • Study shows Yale’s massively popular (and free) “science of well being” course actually does boost well being.

Study: Support the Underground: Characteristics of Beyond-Mainstream Music Listeners
Authors: Dominik Kowald, Peter Muellner, Eva Zangerle, Christine Bauer, Markus Schedl, Elisabeth Lex
Published in: EPJ Data Science
Publication date: March 29, 2021
DOI: 10.1140/epjds/s13688-021-00268-9
Photo: by Joel Muniz on Unsplash