It turns out this is a really hard problem. If you haven't seen it already check out http://www.netflixprize.com/ to set your expectations appropriately.
In Netflix's case it's not that baseline recommendation (whether based on collaborative filtering or something else) is such a hard problem, it's that getting something statistically better than what they have is so hard. Decent recommendation based on CF was a well-understood problem over ten years ago.
Comments
It turns out this is a really hard problem. If you haven't seen it already check out http://www.netflixprize.com/ to set your expectations appropriately.
In Netflix's case it's not that baseline recommendation (whether based on collaborative filtering or something else) is such a hard problem, it's that getting something statistically better than what they have is so hard. Decent recommendation based on CF was a well-understood problem over ten years ago.