First, this doesn't have to be for each pageview. Crudely, you want every story to have a vector in some giant hyperspace characterizing it, and every user to have a vector precomputed offline based on their previously shown affinities, and take dot products.
This is not necessarily easy, or computationally very cheap, but the payoff can be pretty big.
I believe LinkedIn runs something similar to this every night.
Comments
Sounds like a challenge :)
First, this doesn't have to be for each pageview. Crudely, you want every story to have a vector in some giant hyperspace characterizing it, and every user to have a vector precomputed offline based on their previously shown affinities, and take dot products.
This is not necessarily easy, or computationally very cheap, but the payoff can be pretty big.
I believe LinkedIn runs something similar to this every night.
I'm not sure what you mean exactly, but hey, if you can pull it off, it might be successful.
Just think of the number of comments up/down voted every millisecond, and that's just one the dimensions. The scale of that matrix would be enormous.