Last year, Netflix outsourced one of its most difficult technical problem to the public: building a recommendation system for its movie rental service that works 10 percent better than the one it had. A year later, no one has solved the problem, even with a $1 million purse on the line. (For an interview with Jim Bennett, the vice president of recommendation systems at Netflix see “The $1 million Netflix Challenge”.) However, a team of engineers from AT&T has eked out an 8.43 percent improvement, better than any other team and good enough to take home the $50,000 “progress prize.”
The team’s approach was to combine more than 100 recommendation techniques into one mega system that looks for numerous different patterns in the Netflix data–essentially, its customers’ likes and dislikes. The Wall Street Journal has an interview with Bennett about the progress here.