Netflix to shell out $1 million to fix their recommendations
I’d like to take credit for the reason behind the creation of Netflix Prize, but obviously it has been a known issue for quite some time.
Netflix’ recommendation engine just doesn’t work. I’ve written about it before.
The answer is to create the Movie Genome Project (sure its a direct rip-off of Pandora’s Music Genome Project , but still). Perhaps team up with IMDb clickstream data to really tap into what people are interested in. Tie that into the rental history data that they obviously already have and you create the backbone of what should be a pretty powerful recommendation engine.
From Netflix Prize:
The Netflix Prize seeks to substantially improve the accuracy of predictions about how much someone is going to love a movie based on their movie preferences. Improve it enough and you win one (or more) Prizes. Winning the Netflix Prize improves our ability to connect people to the movies they love.
Ok, you have my attention. Interestingly, as the NY Times notes, the type of contest they are running here has its roots in jolly old England:
The prize was modeled on the Longitude Prize, offered by the British government in 1714 to the inventor who could determine a ship’s longitude during transoceanic travel.
Ahhh, web 2.0 meets 1714 England. I love it.
They are looking for a 10% improvement over their current “Cinematch” recommendation engine. Well, that isn’t too hard. Stop recommending that I will like all John Candy movies, just because I enjoyed Summer Rental. Stop telling me that I will like all of the concert videos by a certain band. I do not want to rent concert videos. They really could fix this pretty quickly if they allowed me to tell them a little more about my rental preferences. Allowing me to checkoff what I absolutely do not want to rent will go a long way towards letting the engine recommend actual movies that I want to see.
The problem that I see right off the bat, is they are giving recommendation data out in hopes that someone can build a better moustrap. The current mousetrap is not very good. They are missing the point here by not also giving out rental data. That is where the real work is being done. The recommendation setup of 1-5 stars is to broad and random. Plus, if more folks are like me, the data is meaningless. I try to game the system by purposely rating things in a way that I hope will stop them from recommending movies that I really have no interest in (John Candy example).
Help them out a little and get $50,000 (what they’ll award yearly if no one gets the 10%). Get them to the 10% mark and walk away $1 million richer.
The dawn of consumer-generated-math.
