University of California, Irvine researchers used a computer algorithm that analyzed the properties of more than 500,000 songs released over three decades to extract the most popular ones.
They set the success benchmark for songs as cracking the Top 100 Singles Chart in the U.K. between January 1985 and July 2015, while crowdsourced data from the MetaBrainz Foundation's MusicBrainz and AcousticBrainz projects was used to measure the songs' acoustic attributes, which included the singer's gender and emotional state.
The researchers applied the "random forest" machine learning algorithm to determine the defining characteristics of the songs listed on the chart.
"Successful songs are happier, brighter, more party-like, more danceable, and less sad than most songs," they note.
The algorithm then evaluated 1,052 songs released in 2014, and correctly predicted their popularity 75% of the time, or 85% with the inclusion of artists who had songs charted in the previous five years.
From Los Angeles Times
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found