The Netflix Prize – not what’s needed
I am a big Netflix user. Netflix has a library of about 100,000 movies that users can watch. Because of the size of the library, much of their business comes from customers who have a strong interest in film and want to see movies that they’ve read about in books and are not otherwise easy to find. In fact, Netflix is often talked about as the one of the biggest proofs of the long tail phenomenon in action, along with Amazon.com. Their bread and butter is the semi-obscure.
You have probably heard about the Netflix Prize, which promises a million dollars to the person who programs the best improvement on their Cinematch(SM) recommendation system. I have found a lot of movies that I’ve enjoyed using Netflix’s recommendation system, and have never exactly been disappointed by it. A recommendation from Netflix is only one factor in a decision to have a movie sent to me. Other, more important factors are facts about the film. Their recommendation system already does all the thinking for me that I want it to do. I don’t really need it to be improved. I need it to be easier for me to find what I want based on my own thinking.
What I want from the Netflix website that I’m not getting is a more useful search engine. The search engine on the site searches words in titles and directors’ and actors’ names, and does fuzzy matching to catch misspellings, which is useful. But it doesn’t do anything beyond that. It doesn’t search the synopses of movies, and there is no indexing of the films based on subject matter, or other people involved in the film (composer, screenwriter) or much else. There are user-created lists of movies based on certain themes, but these are difficult to find and tend not to be all that useful.
The reason I want a better search engine is not mainly because I’m a librarian, though that may be part of it. The reason I want a better search engine has to do with the way movies interest me. When I think about movies I want to watch, I don’t think, “Hm, how can I find a movie that I will rate with five stars?” More likely I think something like, “I’m really liking these Robert Altman movies from the 70s, I want to see more, or other 70s movies with Elliott Gould.” That’s actually something that Netflix accomodates just fine, but there are also times when I’m interested in seeing movies with certain subject matter, like movies about con artists or movies about writing. Netflix does not make it easy to find them; other sources are required to pull them together.
There a number of things Netflix could do along these lines, and some of them I think it is already doing. I have noticed in its recommendation pages that narrower genre categories have appeared, like “Pre-20th Century Period Pieces” and “Classic Movie Musicals from the 1930s.” That is certainly a step in the right direction. However, they don’t make it easy to access these narrow genre lists. (They also treat “Foreign” as a genre, which strikes me as ignorant.)
These narrower genres, though, don’t give the ability to search on subject matter, which is often the thing that makes someone choose a movie. For example, if someone in my family is diagnosed with autism, I’d be interested in seeing films that relate to that (of which there have been a number). Similarly, if I’ve just lost my job and have a lot of time on my hands, I might want to see feel-good movies about people who lose everything and then have some kind of good fortune or succeed because of their creativity and persistence. The facts about a movie that might make it relevant can be things like “stock car racing,” or “horses,” but equally, and I think this is something Netflix would have to suggest to users, the theme or type of plot that the movie has, or genres and sub-genres in the industry sense.
What Netflix doesn’t quite get is that their service, with its immense library, implies an approach to marketing that is entirely different than what the studios have to do to get people to come to see movies in current release. It’s not only movie buffs and the intellectually curious who can be better reached with an attribute approach rather than an algorithm (though of course I would like it if Netflix encouraged more people to be intellectually curious). This is because Netflix has the potential to interest people in movies based on changes in their intellectual interests and emotional needs from day to day, which is something that a recommendation engine doesn’t capture since it assumes that people are basically the same one day to the next. The size of the Netflix library and the nature of their service should allow them to do marketing based on more than just the single dimension that the recommendation engine creates.
I think that million dollars needs to go in another direction. So, they should call me.