Item based recommenders can be very effective even with very large numbers
of items, especially if the distribution of usage is a long-tailed one so
that you get significant amounts of co-occurrence.

Utlimately, the sparsity will hurt you, however.  Then you have to move to a
latent variable formulation.  That provides smoothing so that you can reason
from one item to another even without any explicit co-occurrence.  LSI was a
very fashionable choice for this at one time, but there are much better
methods available now.

On Mon, Jan 19, 2009 at 5:31 AM, Sean Owen <[email protected]> wrote:

>
> Do you have relatively lots of users, or lots of items? If you have
> relatively few items, and item-based recommender is ideal -- and vice
> versa with user-based recommenders.




-- 
Ted Dunning, CTO
DeepDyve
4600 Bohannon Drive, Suite 220
Menlo Park, CA 94025
www.deepdyve.com
650-324-0110, ext. 738
858-414-0013 (m)

Reply via email to