Last September, the winners of the ‘Netflix Prize,’ an open contest to develop an improved movie recommendations engine, were announced. I was personally curious about the contest because I had quit the service precisely for the issue that it was trying to solve: I was having trouble discovering new movies to watch via Netflix’s system of recommendations.
I rejoined the service recently to see if there have been noticeable improvements to the recommendations and was disappointed to see lack of progress. As before, the recommendations appeared either irrelevant, unappealing or obvious. An advanced algorithm is hardly necessary to recommend a James Bond film, if I had rated positively other films in the James Bond series. That vast majority of the 31 recommended movies Netflix suggested (based on 370 of my movie ratings) were of the ‘too obvious’ category.
Netflix announced that it will hold Round II of its contest and this time will incorporate additional factors such as “ages, gender, ZIP codes, genre ratings and previously chosen movies,” yet it is hard to believe that this approach will yield results.
Similarly, Amazon.com has struggled with a recommendations engine for much longer than Netflix and is still far from successful.
It seems to me that Google’s approach to social search (which moved to Beta several days ago) is aimed precisely at solving these types of problems. When technology finds a limit, it’s your network that can come to the rescue.
The same principle lies behind the reasoning of why Twitter was hyped as a Google Search competitor several months ago — leveraging the wisdom of your network can beat sophisticated technology. That may not happen outright, but social elements can fill technological gaps.
Even though Netflix does have some community features such as friend suggestions and Facebook connect, they do not seem to be integrated into the larger recommendations engine so the burden to discover new film choices is still on the user. Google’s example of elegant integration of social content into its regular results would serve Netflix well.
Perhaps sometime in the future, a recommendation for me could be slightly less obvious than From Russia with Love.
I think the recommendations on amazon are terrific in non-fiction. I find it difficult to visit their site and not be interested in a purchase. But fiction and movies are much more difficult.
I wonder whether social search in itself will improve movie/fiction ratings. My tastes in that area are different from my friends as well, and even often completely different from my wife. But I have no idea how common that is. What's your experience with that?
But what the previous netflix competition proved was that combining techniques improved results. The teams that met the target both did so by combining the results of lots of teams.
So we can expect rating improvements in milking and combining whatever data or method you can get your hands on, and social search is one part of that. But it is difficult to say how big those improvements will be and whether they will ever work for movies or fiction.
Maybe as a user of netflix you should accept that recommendations are very hard, and that you will get recommendations for movies that you will not like (even my friends do that). But if you watch the recommended movies you will see more (miss less) movies that you actually do like a lot. (like watching your favorite sports team – many matches are just boring, but some matches are so good, you wouldn't want to miss those, so you want to watch every match live). For great experiences you still have to invest a lot of time, but maybe netflix will lessen that time, and show you more gems.
make a contest like this really needs to be focused depending on age, because if we do a mix we ever get the results you're looking for. And even doing it again you can imagine the huge difference in tastes that may exist.