The million dollar prize for improving Netflix’s recommendation system (Cinematch) by 10% has been out there for a couple of years now, and programmers are now within a few tenths of a percent of getting there. The New York Times has a new article about it here. Not a lot of new information over the various articles that have come out since the prize was announced, but I’m still stymied by how weird I am, apparently, in the movie rental world.
Cinematch has, in fact, become a video-store roboclerk: its suggestions now drive a surprising 60 percent of Netflixâ€™s rentals. It also often steers a customerâ€™s attention away from big-grossing hits toward smaller, independent movies. Traditional video stores depend on hits; just-out-of-the-theaters blockbusters account for 80 percent of what they rent. At Netflix, by contrast, 70 percent of what it sends out is from the backlist â€” older movies or small, independent ones. A good recommendation system, in other words, does not merely help people find new stuff. As Netflix has discovered, it also spurs them to consume more stuff.
For Netflix, this is doubly important. Customers pay a flat monthly rate, generally $16.99 (although cheaper plans are available), to check out as many movies as they want. The problem with this business model is that new members often have a couple of dozen movies in mind that they want to see, but after that theyâ€™re not sure what to check out next, and their requests slow. And a customer paying $17 a month for only one movie every month or two is at risk of canceling his subscription; the plan makes financial sense, from a userâ€™s point of view, only if you rent a lot of movies.
Okay, first off, I do want to know how they’re deriving the fact that Cinematch is responsible for 60% of Netflix’s rentals. Are they going off how many people rent movies from the recommendation page or from the “movies like this” thing that pops up when you add a movie to your queue? Because if so, the number’s suspect. To me, saying a recommendation system is responsible for a rental means that the person would not have rented it without the recommendation. But I add things ALL THE TIME from the pop up thing not because I didn’t know about the film or that I wanted to rent it but because it’s convenient and saves me from having to search for it. But that’s a bit by the by.
“New members often have a couple of dozen movies in mind that they want to see, but after that they’re not sure what to check out next.” Wow. I can’t even imagine that. I have somewhere around 1450 films spread across three Netflix queues (you can only have 500 per queue), and that’s not including the 400+ discs worth of TV shows that I have in yet another queue. (It does, however, include several instant watch movies that I’ve seen before and probably wouldn’t rent if they weren’t streaming for free.) A couple of dozen? I’m sorry, I can’t wrap my head around that. There are probably 24 films in theatrical release RIGHT NOW I’d see. Much less the last 100 years of cinematic history. Clearly I am strange.
There’s also an interesting bit earlier in the article about how much Napoleon Dynamite; and other love-them-or-hate-them films like Lost in Translation; and I Heart Huckabees; throw off the system, because it’s so difficult to predict whether someone will like them or not. I can totally see that, and all the films they mentioned are ones that I tend to avoid recommending most of the time, for the same reason. Except Lost in Translation, because I have mad, blind love for Sofia Coppola.
The good thing about all this is, I think, the fact that the recommendations are apparently encouraging people to check out more offbeat, older, and independent films. That’s a great thing about a subscription service with so wide a selection – the cost of experimentation is very low. I often think this should be a critic’s job, too – rather than warn people away from the latest multiplex blockbuster that they’re going to see anyway, turn them on to a hidden gem they might otherwise miss amidst the flurry of big studio publicity. (I think Netflix should put up pages for all the major festivals, since that’s where the best indies first come to light. It would certainly save me a lot of time and effort currently spent in searching Netflix for every festival film every few months in case it suddenly ended up with a distribution deal. /selfish)
So tell me, do you use Netflix’s recommendations? Or if you’re not a Netflix subscriber, some other sort of algorithmic recommenations, like Flixster? Does it influence 60% of your rentals? Does anyone relyÂ on recommendations of this sort, rather than also factoring in human recommendations, whether from friends, critics, or bloggers – or a personal affinity for a cast or crew member? In other words, if Netflix recommended a movie (with a higher than 4.5 predicted star rating, let’s say) you hadn’t heard of, and you didn’t know any of the actors or the director, would you rent it without digging up more info? I wouldn’t. But as already decided, I am strange. And maybe I would get to that unknown film, once I got done with the 1400 already in my queue. :)
Ooh, just got to the end of the article (yes, I’m reading and writing at the same time – sue me), and found this: “[Netflix CEO Reed] Hastings is even considering hiring cinephiles to watch all 100,000 movies in the Netflix library and write up, by hand, pages of adjectives describing each movie, a cloud of tags that would offer a subjective view of what makes films similar or dissimilar. It might imbue Cinematch with more unpredictable, humanlike intelligence.” I WANT THAT JOB. When that job is posted, Mr. Hastings, let me know, mmkay?