Any ideas about how to improve collaborative filtering? Win $1 million
2 Oct 2006 - 4:00am
I just read an interesting article on news.com about a competition
that Netflix is starting. They want to improve their recommendation
engine and they have tapped out the skills of their internal team. To
that end, they are offering $1 million to someone who can improve
their system by at least 10 percent.
I have only superficially been involved in such systems. But, it
appears tha Netflix is looking for an algorithm to solve the problem.
Shouldn't there be more of a human-factor approach, rather than a
brute-force approach to this issue? Have they exhausted all of the
behavioral research, cognitive science and psychology approaches?
Even moving pictures wouldn't have been possible without
understanding how the brain works.
If we analyze how/why people give recommendations and then look at
what makes a recommendation useful, then we can start developing
creative solutions to bridging this gap. Obviously, there is a great
difficulty in that what we like or dislike changes based on our
bodies chemical balance, our level of sleep deprivation, and other
external factors. That is why some movies, music or art are only
years later appreciated.