Crystal balls would probably be a real hit with more than mere wizards. Financial advisers, economic forecasters and investment bankers would certainly be high up on the list of those in desperate need of one right now, but they might also come in handy for a large spectrum of other professionals. Like record company executives whose careers heavily depend upon their ability to discover the next big star. Prof. Yuval Shavitt of Tel Aviv University is hoping to give those record companies a hearty scientific boost to fortify their gut instincts. By harnessing the power of prediction in a reliable way, he is able to do the work of crystal balls and accurately divine the future. Using software instead of sorcery, Shavitt and two graduate students, Tomer Tankel and Noam Koenigstein, have developed an algorithm based on data collected from Gnutella, one of the most popular and largest peer-to-peer file-sharing networks in the United States. Over a nine-month period in 2007, the team developed a system to predict whether or not an unknown artist would go from local popularity to national fame. They started by examining the first six months of data and then used the remaining three months of data to track the increasing popularity of unknown artists. According to Shavitt, 70 percent of the queries in Gnutella are people looking for music and about half of the users are American. "Music data has short strings, like the name of a singer or song or even a genre, such as hip hop," says Shavitt. "We analyzed these strings in different ways and realized that there was a correlation between what was happening in the network and real life." In cases like Madonna and Britney Spears, well-oiled publicity machines make their release of a new song or album known almost instantly all around the globe. In cases of hugely popular artists like this, the network serves more as a mirror of real world events and knowledge. But when it comes to young, unknown artists, spikes in popularity on the network in a specific geographical region can indicate future success. "We see a definite correlation between the number of queries for a particular artist and what happens to them later," Shavitt explains. "We look at the demand for their songs and if it grows rapidly in one geographical location, others usually follow and then we often see these artists getting contracts and appearing in Billboard." AS SHAVITT points out, the same algorithm is nearly impossible to track on search engines like Google and Yahoo, because people could be searching for anything and it would be extremely difficult to differentiate songs and artists from everything else. Alternatively, if an artist or song is growing quickly on the network, it's easy to see and define it. In most cases, the software has about a 30 percent accuracy rate, but in some locations, such as Atlanta, that figure jumps to 50%. The explanation for this is that Atlanta is the home of hip hop, so if a hip hop song spikes in popularity there, the chances that it will later go national are even greater. The team defines success in two separate ways that are often correlated. One indication is when an artist becomes popular across the entire nation on Gnutella. Another is if they reach the Billboard charts through album sales and radio station play time. "If they make it into one of the top 100 most played songs in Billboard, they're successful," says Shavitt. "That's not easy to do." So far, the software has successfully predicted the future success of numerous artists. In April 2007, it flagged two artists' songs weeks before they became nationally popular and eventually hit the Billboard charts: Soulja Boy's "Crank That" and Sean Kingston's "Temperature." Their biggest software achievement to date came with the group Shop Boyz. "Nine weeks after we spotted them, they signed a record deal and three weeks after that they were in second place on the Billboard charts," explains Shavitt. "Exponential growth is a strong signal about what will happen in the future." Yung Bird, Huey, Hurricane Chris, Cupid, and Mistah F.A.B. are other examples of groups spotted by their software before making it on the national level. Yet, while other software prediction programs do exist, most of them require listening to the music itself and analyzing it to give an opinion about its future fate. Shavitt's software is far more efficient as it only examines numerical data from Gnutella, between 15 million and 40 million queries a day. But the critical understanding that led to successful predictions was the relationship between the geographical location and rising popularity. Artists who make it to the national level often start out with moderate success at home that grows rapidly. In some cases, as few as five queries a week then jumped to 20 and 30 and finally 150 within a specific region. If Madonna hadn't been spotted by a talent scout in New York and Britney Spears hadn't signed with a record label in Louisiana, they may never have made it onto the national scene. Shavitt hopes his software will change that for other artists. It's very difficult to garner attention these days, but if unknown artists have a local following that grows, they may be able to make a case for record deals. The applications for record companies are obvious, but Shavitt says that the software could also be used to predict the future success of young, unknown politicians and any other trends that begin locally and then go national. Currently contacting record companies, Shavitt hopes to either sell the software or form a company that would provide results for a fee. "We don't have a good grip yet on all the opportunities, but we're working on it." For now anyway, the crystal ball is under tight supervision.