Hacker News new | ask | show | jobs
by psionski 4818 days ago
The article didn't say, does this site have any intelligence? A modern porn site would be like Netflix, i.e. it will slowly "learn" what (sub)genres you like, whether you like blondes, brunettes, etc. and it will automatically suggest videos to you.
1 comments
hollerith 4817 days ago
In my experience, Netflix's recommendations are approximately useless. Is your personal experience, or the experiences of people you know, different than mine?
link
freehunter 4817 days ago
When I was single and watching Netflix by myself, I thought their suggestions were good. I put a lot of time into rating movies and shows I've already seen, and it paid off in the suggestions. Once I moved in with my SO and we started sharing a Netflix account, the suggestions became useless. Some days she watches what she wants and some days I watch what I want. Other times we're watching something together that we both enjoy. The problem is, Netflix can't handle all of this, so we end up with some rather odd suggestions (how in the world does Netflix think I care enough about Russia's Toughest Prisons to make it one of my top 10, right alongside Visoneers and Extreme Couponing?)
link
rquantz 4817 days ago
I've been wondering why Netflix doesn't come up with some kind of personas. Each member of the household can rate movies, and then it can make recommendations for an individual or combination of them.
link
psionski 4817 days ago
Is there a way to compute a "taste distance" for a combination of users reliably? I think if you just throw their ratings together you'll get a point with impossible taste (except if they like the same movies, in which case it won't matter). Maybe return only movies that were in the recommendations of both users?
link
hollerith 4817 days ago
>Once I moved in with my SO and we started sharing a Netflix account, the suggestions became useless.

That explains it then. Much of my account's viewing history is stuff watched with the SO. Apparently the recommendation engine puts a lot of weight on this viewing history. (Until today, I had assumed that recommendations depend only on those 1-5-star ratings.)

link