I am running average absolute difference evaluations of a generic user based recommender that uses a threshold based neighborhood and pearson correlation to determine similarity.

I evaluated several recommenders for varying minimum thresholds for the neighborhood (0.9, 0.8, 0.7, 0.6, 0.5)

I noticed that as I decrease the threshold, the average absolute difference actually goes down, from:
0.85299 difference at 0.9 threshold of similarity
to
0.77667 difference at 0.5 threshold of similarity

My original intuition was that a higher threshold of similarity should result in more similar users appearing in each neighborhood, and hence should result in lower average absolute differences. However, this does not appear to be the case. Is there possibly some theoretical reason behind this? I repeated the same experiments using uncentered cosine similarity and those results reflect my original intuition (decreased difference when minimum thresholds for neighborhoods are higher)

I am performing experiments over the movie ratings from group lens.

True. But that neighborhood of more-similar users is smaller, and the
things they know about are fewer. Maybe better recommendations are found
attached to users a bit farther away, that are not recommendable when
considering only a very small and close neighborhood.

On Wed, May 9, 2012 at 5:13 PM, Daniel Quach <[EMAIL PROTECTED]> wrote:

> I am running average absolute difference evaluations of a generic user
> based recommender that uses a threshold based neighborhood and pearson
> correlation to determine similarity.
>
> I evaluated several recommenders for varying minimum thresholds for the
> neighborhood (0.9, 0.8, 0.7, 0.6, 0.5)
>
> I noticed that as I decrease the threshold, the average absolute
> difference actually goes down, from:
> 0.85299 difference at 0.9 threshold of similarity
> to
> 0.77667 difference at 0.5 threshold of similarity
>
> My original intuition was that a higher threshold of similarity should
> result in more similar users appearing in each neighborhood, and hence
> should result in lower average absolute differences. However, this does not
> appear to be the case. Is there possibly some theoretical reason behind
> this? I repeated the same experiments using uncentered cosine similarity
> and those results reflect my original intuition (decreased difference when
> minimum thresholds for neighborhoods are higher)
>
> I am performing experiments over the movie ratings from group lens.