If you like math or movies or human psychology, among other things, I’d highly recommend this particularly interesting article of the NY Times, titled If You Liked This, Sure to Love That. It details at Netflix contest for individuals to improve upon their current program of movie recommendations such that it works 10% better than it already does (in the sense that it can predict 10% more closely how a viewer will like or rate a particular movie). The article is nice, descriptive, in depth look at many ideas, so it might take you 15-20 minutes to read, but I wanted to delve into a few of the issues it brings out that I was intrigued by.
The article starts off describing the “Napoleon Dynomite” problem — basically that it’s extremely challenging to predict based on movie taste and past movie ratings whether or not a person will like this movie. I, myself, still have yet to see this movie, but I definitely know people who loved and hated it, and as the article mentions that the ratings for “Napoleon Dynomite” are disproportionately 1 or 5 stars (the highest and lowest possible for the Netfilx scale).
This issue couples with another idea question brought up of whether a computer can do better at making recommendations than a human. While the computer has tons and tons of data at it’s “hard drive” tips, to mangle a phrase, there is something about the human perception that does a pretty good job at discerning likes and dislikes of another person, even if the person doing the perceiving is the clerk at the local DVD store (we’re past video stores now, yes?). The article mentions, too, that a computer is more likely to play it safe while a person draws upon their own likes and dislikes as well and may go a bit more on a limb that could be much more accurate than a computer, but may also come up short more regularly. So the question becomes, “Can anyone’s enjoyment level truly be determined based on their previous levels of enjoyment of similar and dissimilar events?” And if so, would we be doing ourselves a disservice to never experience things that might actually cause us dis-enjoyment? Isn’t it good to experience both?
The other piece of the article I liked was the math – and it’s one of the reasons I might recommend it to someone :) It was interesting to read how different algorithms were used and combined to do the math of movie suggestions. Even more interesting was reading that as things got more or more complex, even those writing the computer programs no longer really recognized what the program they had written exactly was doing, but just knew that it seemed to be working than the one that preceded it! To me, it’s curious that we throw in some data to an algorithm we really don’t understand and receive back a satisfactory answer that then can be tested for accuracy and reliability, but in between we’ve lost sight of what’s happening.
Is an answer worth getting if you don’t know how you got it? The math teacher in me says no, but the movie lover in me doesn’t care as much. If you’re able to take me love for “Hoop Dreams,” “A Clockwork Orange,” and “10 Things I Hate About You” and provide me with an enjoyable way to spend a Saturday evening that I wouldn’t have found on my own, I just might take it. Or maybe instead I’ll just take it up with my friends.