Yes, you're trying to solve a slightly simpler problem than
recommending items to users. You just want to find users similar to a
given user.

Yes you di want the method UserBasedRecommender.mostSimilarUserIDs().
It's implemented at the moment by GenericUserBasedRecommender and
others.

An item-based recommender is based on item-item similarities, so it's
not by nature suited to answer the question, which users are similar
to a given user? It's user-based recommenders that can help you there.

I don't think you want to swap users and items, no.

Sean

On Mon, Mar 1, 2010 at 9:53 PM, Steve McNewman <[email protected]> wrote:
>
> Hi Mahout Community,
> I am currently playing around with the Taste and would like to get the
> following working: Given a set of users who have used a certain item, I
> would like to get a list of recommendations for the most similar users. As
> far as I understand, both user- and item-based recommender return a list of
> RecommendedItems when executing their recommend(userId, howMany) method. The
> difference between them is that user-based recommender use (weighted)
> user-similarities and item-based recommender (weighted) item-similarities,
> right?
>
> How do I get the most similar user IDs? Do I simply use the
> mostSimilarUserIDs(userId, howMany) method of the user-based recommender?
> Why can't I use an item-based recommender for this? Do I need to use a
> DataModel with swapped item and userIdColumns? In how far is the list of
> most similar User IDs related to collaborative filtering?

Reply via email to