Monday, November 24 
Good article in the New York Times about Netflix’ competition it’s been running the last couple of years to improve its recommendation engine. Hundreds of people have tried to win the $1 million on offer, but no-one has yet cracked the 10% improvement barrier that Netflix has set. In other words, the contestants have to improve the quality of suggestions that Netflix makes to its members based on their previous ratings by 10%.
There are many reasons why this is difficult to do, attested by the asymptotically slowing rate of improvement. One class of movies is particularly tricky - those which are love-it-or-hate-it in nature. Napoleon Dynamite epitomizes this category - Netflix and the contestants have a terrible time predicting whether someone will like it or not based on their past track record.
People go through phases in buying things, and our multiple personalities come out in what we buy and enjoy. We are not (entirely) self-consistent, logical beings.
Even so, it sometimes seems like recommendation engines just seem to miss the boat. By all rights, Amazon should be able to make highly accurate recommendations to me about all sorts of things, given the amount of stuff I’ve bought there, yet it is incredibly inaccurate. It keeps recommending products that I’ve already bought (e.g. I just bought one digital camera so it immediately recommends another - why would I need two?) or which are off-base in the long term (I buy a gift for a baby shower, and it throws all its reco’s off).
ny times, amazon, netflix, recommendation engine, social software, search, clive thompson | IN CATEGORIES... Technology, User Experience, Software, Innovation