The science behind chart success

Summary

Researchers at the University of Antwerp have examined exactly what makes a dance record likely to be a hit

Mining for hits

What makes a song a hit? It’s the million-dollar question all record labels would love the answer to. Researchers at Antwerp University have now developed an online application that can predict the hit potential of dance songs.

The co-ordinator of the research project was Dorien Herremans, who carried it out as part of her PhD. “We know that record companies in the US use certain data mining techniques to examine a song’s potential, but they don’t disclose any information on it,” she says. “Earlier academic projects have largely failed because they didn’t concentrate on a specific type of music, which we decided to do.”

The focus on dance music meant the researchers could clearly define which data to use and which songs could be considered hits. They set up a database of about 760 songs, from 1985 to 2013, which were listed in the Billboard charts and by the Official Charts Company. A song was considered a hit when it entered the chart’s top 10.

The researchers then examined in detail the songs’ basic audio characteristics, such as the tempo, defined by the number of beats per minute, and the volume. One of the findings of this analysis, which may not sound too surprising, was a clear trend towards increasing tempo and volume in dance music.

With their data, the researchers created an online application, which determines how likely it is that a song will become a hit based on the different audio characteristics. “A test on about 75 songs showed that the model works with about 83% accuracy,” says Herremans. The application can be used for free by anyone who wants to upload a song to test its hit potential.

Media attention has helped to create a lot of interest in the application, which has been used about 900 times in just a week. However, Herremans acknowledges that it is not the Holy Grail of the dance music business.

“Important characteristics of songs have not been taken into account for the development of the model, such as lyrics and the artist’s popularity,” she says. She hopes to integrate such additional characteristics in the future, as part of a possible follow-up project.

About the author

No comments

Add comment

Log in or register to post comments