I know you're thinking along the lines of matching items to preference
settings, but assuming for the sake of argument that there were
actually people using the service, I think one relevant area could
also be identifying people who starred a similar set of items (each
vector is a persons set of starred items). You could also perhaps
pick out items that are similar to each other based on a similar set
of people starring them (each vector is the set of people who starred
a particular item)?
- Luke
On Jun 19, 2008, at 4:09 PM, Randall Leeds wrote:
http://www.mit.edu/~andoni/LSH/
To ponder:
What near-neighbor searches do we do in Melkjug?
Which might this apply to best?
Maybe we should just implement some of them and see how it goes?
(I'm actually doing this...)
-Randall
!DSPAM:4013,485abd65171202458217002!
