Netflix Prize competitors join forces, cross magic 10-percent mark

A coalition of four teams representing a quartet of countries looks poised to collect a $1 million prize after managing to move the needle on Netflix's recommendation engine by 10.05 percent.

Wired Epicenter blogger Eliot Van Buskirk (a former colleague of mine at CNET) reports that BellKor's Pragmatic Chaos, a combination of four teams vying for the elusive Netflix Prize, submitted a new "solution" to Netflix on Friday that would improve Netflix's video recommendations by more than 10 percent—enough to snag the $1-mllion prize that's been up for grabs since February 2006.

The four teams—the two-member Pragmatic Theory of Montreal (which, on its own, had managed to reach the 9.80-percent mark), a pair of AT&T researchers in the U.S., a research scientist for Yahoo! Research Israel (go, Yahoo!), and two "machine learning" researchers from Commendo Research and Consulting in Austria—had been circling the Netflix Prize for some time now.

But as one member of Pragmatic Theory told the New York Times: "Because of the nature of the competition, making a collation of teams is a quick way to improve results."

Team BellKor's Pragmatic Chaos won't be able to collect the massive prize right away, however. Van Buskirk notes that first, other Netflix Prize competitors have 30 days to match or exceed the would-be winner's progress. If, after that, Bellkor's Pragmatic Chaos is still in the lead, Netflix will go ahead and verify the results. If the team's submission passes muster, the seven researchers will split the big prize.

It's been a long, suspenseful wait for a Netflix Prize winner. The various competitors—ranging from award-winning researcher scientists to, literally, "Just a guy in a garage"—made progress in fits and starts, and the closer they got to the 10-percent finish line, the slower their progress got. Indeed, as one of the winner-elects lamented to Wired: "For a long time, we weren't sure if 10 percent was even achievable."

But will all that work result into better movie recommendations from Netflix? The movies-by-mail giant already tweaked its recommendation engine back in May using some of the formulas developed by Netflix Prize contestants, and as I wrote at the time, my personal picks have been pretty accurate, give or take. Hopefully, we'll get the answer in about 30 days.

Correction: In the original version of this post, I incorrectly reported the percentage by which team BellKor had improved Netflix's recommendation engine; the correct figure is 10.05 percent, not 10.5 percent. Apologies for the goof.

Related:
Winning Teams Join to Qualify for $1 Million Netflix Prize [Wired Epicenter]