Re: K-Means

[Subbarao Kambhampati <rao@asu.edu>]

The distance is application dependent. For the present case, distance
should be seen as inverse of similarity.

rao


At 12:40 PM 5/9/2001 -0700, you wrote:
> Hi,
>  
>     I have a question on the
> fifth problem.  With the k-means algorithm, each iteration the new
> cluster center is the item that is closest to the mean of the
> cluster.  What is not clear to me is how is distance from the
> cluster center defined for each document.  Is it Euclidean distance,
> or the similarity between a document and the mean cluster center, or some
> other metric?
>  
> --Thank you
> Jeremy