Little known fact, but we did start out with the goal of amassing a large library, and we even put significant effort with some former Netflix-prize winners in building a recommendation engine.
The first big problem, which you see on Netflix as well, is that a recommendation engine really falls off in efficiency when your selection is thin, and the original dream of streaming every film is never going to happen because of a price squeeze from the content side (hence Netflix moving into production and focusing on exclusive content).
The second, perhaps bigger problem (particularly for our audience) is that there is only so much metadata that you can pump into a film database that will tell you categorically if someone wants to watch it. People like to be delighted by discovering a new film, not because it is somehow akin to 20 other films they liked, but because of where they are in their life now, and the contemporary meaning of that film. I know curation has become possibly the most annoying buzzword in SV circles lately, but it has been our approach since 2007, real actual curation by human beings—not algorithms that purport to be "curatorial" on a pitch deck.
Regarding your second point, I really understand this position - fair enough (re. zeitgeist and personal point in life of someone looking for films and rating films, etc.) I just wonder if it wouldn't be interesting to spend some time looking at recommendation systems which are not just simply about intersecting users' votes and presenting something "just what you've recently liked." But this would be more of a playful experimentation which is more hobbyist in nature, for sure.
What, so like, "as an early thirty-something who just found out you're about to become a father, you may enjoy films X, Y, Z"? To really get people's life chapters and such you'd have to have massive metadata about them beyond the scope of a video service. Google/FB could do it, though.
It was more a comment about how considerations such as this make collaborative recommendation systems precisely less useful. Regarding rec.systems that move beyond naive rating/like intersections between users (such as e.g. Jaccard similarity index), I think there could be interesting algorithms developed that make use of assigned/curated item tags together with collaborative filtering/recommendations (e.g. something that last.fm has considered (or has been considering) - there's an entire field of research for just music recommendation systems.)
I would pay $10 a month for a recommendation engine only. I'm able to find the actual video files easily enough. (And I often do so, even with Netflix has it, as Netflix's player sucks, quality is iffy, subtitles are spotty, etc.)
The first big problem, which you see on Netflix as well, is that a recommendation engine really falls off in efficiency when your selection is thin, and the original dream of streaming every film is never going to happen because of a price squeeze from the content side (hence Netflix moving into production and focusing on exclusive content).
The second, perhaps bigger problem (particularly for our audience) is that there is only so much metadata that you can pump into a film database that will tell you categorically if someone wants to watch it. People like to be delighted by discovering a new film, not because it is somehow akin to 20 other films they liked, but because of where they are in their life now, and the contemporary meaning of that film. I know curation has become possibly the most annoying buzzword in SV circles lately, but it has been our approach since 2007, real actual curation by human beings—not algorithms that purport to be "curatorial" on a pitch deck.