Re: Scalar Clustering

[Subbarao Kambhampati <rao@asu.edu>]

I am not sure I understand your question. In the example given, you will 
find the most related terms to the terms used in the query using local 
clustering analysis, and also elaborate the query and see how the results 
change with the elaborated query.

So, suppose the original query was "data"

you get some results. Take top 10. Do analysis, and find that "structure" 
and "algorithm" are the two closest.

Assuming you are using 2 closest to elaborate the query, your new query is

"data structure algorithm"

Now you will get a new ranking.

To evaluate the effecitiveness of clustering, you compare the new ranking 
to the original ranking (with just "data"), acting as if you are a 
disinterested user.

Rao


At 04:31 PM 4/1/2001 -0700, you wrote:
>I though we were supposed to find the most related terms to the terms used 
>in the query?  Are we suppose to suggest the most realated terms that 
>appear?  I.e. suggest structure and algorithm?
>
>
>On Sun, Apr 01, 2001 at 04:17:21PM -0700, Subbarao Kambhampati wrote:
> > At 02:48 PM 4/1/2001 -0700, A wrote:
> > >     I had question concerning scalar clustering and the project.  Would
> > > it not be better to use associative vectors to determine revelant terms
> > > when you only have one word in a query.  Because when you use scalar
> > > cluster vectors and there is only word in the query you either get a 
> 1 or
> > > 0, either a word is related or it is not.  At least with the associative
> > > vectors you get some value between 0 and 1 and from this you could
> > > determine how related the term is.  I don't see the advantage of using
> > > scalar clustering over associative.
> >
> > Actually, the number of query words has no impact on whether you should do
> > scalar or associative clustering. Consider the query "Data". It may 
> well be
> > that by doing clustering on the set of top few docs that were returned for
> > the query "data", we figure out that
> > "structure" and "Algorithm" are correlated with this query.
> >
> > Rao
> >