Algorithm spots music’s next great superstarDecember 9th, 2008 - 3:44 pm ICT by IANS
London, Dec 9 (IANS) For every rock star who makes it big, there are thousands who are lost to failure and anonymity. Before Madonna was “Madonna”, she was a local success in New York clubs. Until Britney Spears became a global pop superstar, she performed in dance revues in her native Louisiana.
But how can you tell who will make it onto the Billboard charts and who will never get beyond the local bar circuit?
Yuval Shavitt, a professor at Tel Aviv University’s (TAU) School of Electrical Engineering, has developed software that can accurately predict the next big music phenomenon.
Using data collected from Gnutella, the most popular peer-to-peer file-sharing network in the US, Shavitt has developed a computer algorithm that can spot an emerging artiste several weeks or months before national success hits.
“Until now, talent scouts for record companies used instinct to predict the next rock personality. Our software has an astonishing success rate, about 30 percent and in some cases up to 50 percent. We’ve crossed a new frontier in the record business,” he says.
Soulja Boy (”Crank That”) and Sean Kingston (”Temperature”) were both flagged by Shavitt’s system in April 2007, weeks before they emerged into the national spotlight, both songs became Billboard hits when they entered the charts in June of that year, according to a TAU release.
And the group Shop Boyz skyrocketed to popularity in their home city of Atlanta in just two weeks. Their “Party Like a Rockstar” became a hit single, and Shop Boyz was catapulted to national fame.
But not before the band popped up on TAU’s algorithm “radar” a few weeks before they signed with Universal.
To develop the algorithm, Shavitt, with graduate students Tomer Tankel and Noam Koenigstein, examined a large amount of data from Gnutella user queries for unknown artists over a nine-month period during 2007.
By examining the first six months’ worth of data, and then using the remaining three months’ data to track the increasing popularity of those artists, they developed a system to predict which artists would break out of their local markets.
“The key was understanding the role of geography in the rising popularity of these artists,” says Shavitt. As part of the largest study ever done on geographically-tied searches, TAU researchers examined the 30 to 40 million queries that are entered each day on Gnutella.
They realised that those artists who eventually made it to the national level first had a huge number of user queries in their local region, even when they had zero queries from elsewhere in US.
This same software can be applied to TV programmes, video clips, and other entertainment products, including home videos on sites like YouTube.