Another thing that is very helpful is simple jittering. The anti-clustering can also work based on discrete known properties. If recommending text, you can just limit the number of items from the same source, for instance.
On Sun, Jun 21, 2009 at 6:27 PM, Sean Owen <[email protected]> wrote: > I see so it's more like picking some most-orthogonal subset of the > vectors actually. I get it in principle I think. I also imagine the > simple "anti-clustering" approach you mention works pretty well. > > On Sun, Jun 21, 2009 at 9:23 PM, Ted Dunning<[email protected]> wrote: > > Gram-Schmidt doesn't have to change vectors. You can view it as a way of > > selecting from an infinite number of vectors in order to get an > orthornormal > > basis. The task of getting an interestingly diverse set of > recommendations > > is a bit different in that we only have a finite number of items to > > recommend and in that orthonormality isn't really a concept. Another way > to > > look at it as greedy set-cover. >
