> > Friday, June 9, 2006, 7:25:10 AM, you wrote:
> >
> > WL> I think we need to *try* to ensure that the results page gives
> > WL> *some*
> > WL> clue(s) as to how we went from your precisely spelled
> > search values
> > WL> to our finely crafted result set, especially when the part in
> > WL> between is otherwise indistinguishable from magic.
> >
> > Why not follow the Google procedure, where if you type in a word like
> > piza it asks 'did you mean pizza'. Of course that may not solve it
> > if you really wanted Pisa.
>
> I agree that following the Google model may be useful, but thinking
> in the context of a faceted model, isn't "Did you mean..." just a
> different facet to be presented to the user along with the facets
> found for the given query?
Thinking about how we handle misspellings and low-result, no-result, or
other types of results sets is valuable, important work (not to mention
fun). But the real proof is in the pudding: test it out. You'll notice that
Google itself keeps changing its configuration (my biggest argument against
the Google Appliance as a search tool, aside from the "secret soup" issue
that you don't know enough about how it works, is that they have a Father
Knows Best approach to its functionality based on the requirements of a
massive full-text database that may not be well designed for your data).
Look at stemming-off, and on, and off, and currently on. Yet if you test
your data with user queries, you may find stemming (not fuzzy match! Oh lord
please not) is essential for decent retrieval.
Karen G. Schneider
kgs_at_bluehighways.com
Received on Fri Jun 09 2006 - 13:21:39 EDT