So, I subscribed to Netflix, and started queuing up things to watch. Another business model constraint became evident when I realized that many interesting movies were only available for DVD's "by mail" as opposed to online streaming -- no doubt a point of conflict between the online Netflix model and the copyright holder. At some point the 'big guys' will get it, and we will see the flood gates open to online video streams from the proper rights holders, but until then we will have hurdles to hop, inconsistencies and inefficiencies.
Netflix is attempting to get to the next generation of "Personalization" -- able to anticipate what you will enjoy using a formula that combines what you have watched (and ideally rated), viewing preferences you explicitly enter, and the preferences of "viewers like you". You may also be aware that they had a contest to see if someone can come up with a better algorithmic way to select recommendations. Given all of this, I was a bit disappointed in the recommendation service circa 2011: indicating that my indicated "like" for "Wallace and Gromit" was the basis for selecting a number of recommendations including "Lewis and Clark" --- sort of slipping between animation/comedy and historical documentaries ... I think that one was avoidable. Mind you historical documentaries are high on my list of preferences and 'likes', so it is not the recommendation that is in error as much as the rationale presented.
A second realization emerged as I started to consider the categorization preferences. First, you only get 5 points of rating, and "ok" is not one of them - you must either "like it" or "didn't like it" as an option. Second, for user preference indicators you only get three levels (never watch, sometimes, and always) which is very weak as well. I sense the classic engineering-marketing concern with the limited supply of positive integers. A sliding scale, encoded between negative 10,000 and positive 10,000 would be easily captured in the same database -- and while a user might not be consistent, it would provide a much more finely tuned level of feedback.
Then there is the basis for selection -- my wife has some series she enjoys (CSI and such), my grandson is currently into sunken ships and submarines, and movies we watch with this part of the family (Disney/Studio Ghibli) don't match the films my wife and I watch together. The result is a mix of selections and preferences which not really a match for any one person -- personalization that lacks personalization.
Finally there is the question of privacy. One of the few legal protections in the U.S. relates to video rentals. This is a result of the records of such rentals being a factor in the rejection of Supreme Court nominee Robert Bork. It is not clear if this applies to Netflix and their personalization. Your selections and stated preferences may disclose things about you which you do not wish to have public. There is a school of "privacy is dead" -- including a recent piece on this in IEEE Computer Society's Internet Computing by Jakob Ericsson. But I for one advocate for some level of guidelines, not quite willing to submit to the ubiquitous audio/video surveillance suggested by Niel Stephenson in Snow Crash with the virtual "earth" -- which apparently inspired "Google Earth" - albeit limited in it's real-time presentation of close-up audio/video streams - particularly indoors.
The bottom line is that the Netflix personalization inspires both fear (implied personal characteristics) and fun (off-target recommendations). We need to develop the policies and principles that will guide this evolution -- even if we cannot do this as quickly as the technology is evolving.