Algorithm to find the next star in music

Researchers at Tel Aviv University are using file sharing networks to find promising artists

Hip-hop music is fueled by young people growing up in Atlanta, New York, Chicago, Los Angeles and quickly rising to the top. But the path of a young musician in the USA is not easy. She starts performing in clubs in major cities in the USA in the hope that he will be found and signed by the major music distribution companies and from there the success depends only on him. But how will the music companies locate the artists? Researchers from Tel Aviv University were involved in this task.

The researchers, headed by Prof. Yuval Shavit from the School of Electrical Engineering at Tel Aviv University, analyze the file sharing networks and based on the analysis of the users' preferences predict with an accuracy of 30-50% the success of the artist, and this without hearing the songs themselves. The research is part of Noam Koenigstein's master's thesis for which he won an award from the Faculty of Management of Tel Aviv University for technological research with commercial potential, and indeed commercial interest has already been expressed in the software developed within the framework.

The software identified the artists Soulja Boy ("Crank That") and Sean Kingston ("Temperature") as early as April 2007 weeks before they reached the Billboard sales chart (the hit list of the American music industry). The Shop Boyz band was identified as having the potential for success as early as February 2007, two months later the band was signed by Universal Republic, a leading music distribution company, and a few weeks later the Shop Boyz put their grown-up hit Party like a Rockstar on the Billboard charts, a few weeks later it reached The song went to second place in the chart.

To develop the software, the team from Tel Aviv University, Prof. Yuval Shavit and research students Noam Koenigstein and Tomer Tankel, tested over half a billion hash strings (queries) that were sent to the Gnutella network to share files for 10 months between the end of 2006 and the end of 2007. The results of the algorithm were compared to the billboard list. This is an unprecedented amount of queries in the study of file sharing networks, every day 10 to 40 million queries were collected, and analyzed in a reasonable time required the solution of difficult technical hurdles.

"The key to success was understanding how the geographical dispersion of the queries indicates the success potential of the artists," says Prof. Shavit, "We realized that artists who succeeded in a big way, did so quickly in the area where they started their career. The geographic segmentation of the queries allowed us to identify successful artists in a certain region when in the US as a whole they were still unknown." The algorithm makes it possible to identify a trend of success even when an artisan receives only about a hundred queries a day, a negligible number in a sea of ​​millions of daily queries. "We are looking for the big trend and not the absolute numbers" says Prof. Yuval Shavit.

For the purpose of continuing the research, the team is now building a new collection system from a smaller file sharing network, direct connect, and manages to collect between a million and a half million queries per day. In the future, the team hopes to succeed and predict the longevity of a song at the top.