Software Predicts Future Rock Stars

The best talent scouts have a special instinct that tells them if a new band has the stuff to make it big - and make the record label money. Now, researchers have developed a software program that uses data from search queries on peer-to-peer file-sharing networks to predict which new artists will have hit songs.

The researchers, led by Yuval Shavitt of Tel Aviv University, hope that music producers and record labels could use the tool to identify the next big music phenomenon. The software might also interest young people who want to be up on the next big trend.

The computer algorithm has shown to be successful so far, achieving a success rate of up to 50 percent (half of the algorithm's predictions turned out to be true hits). For example, the software flagged Soulja Boy's "Crank That" and Sean Kingston's "Temperature" in April 2007, and both songs became Billboard hits in June 2007. The software also identified Shop Boyz' single "Party Like a Rockstar" before the band signed to Universal, and several weeks before the song became a national hit.

To do this, the researchers analyze 30-40 million search queries entered on Gnutella, the largest peer-to-peer file-sharing network in the US. In the beginning, queries almost always come from a single geographic location (the artist's hometown), but if they grow exponentially, the search queries prove to be a reliable predictor of a future hit artist.

Besides serving as a magic ball for record labels, the software could have other applications. Another researcher on the project, grad student Noam Koenigstein, hopes to use the software to examine new songs by already popular artists like Madonna, and see if these songs sell well because they have the criteria of a "hit," or just because they're sung by Madonna. In addition, the researchers said that the same software can be used to predict the popularity of other entertainment products such as TV shows and YouTube videos.

via: Tel Aviv University