[WEB4LIB] Re: metadata for site searches

[Thomas Edelblute]

I have to disagree.  Free text searching yields too many false drops, often over
50%.  I am a firm believer in using a controlled vocabulary with adequate
cross-references.  I would recommend a full authority file to keep track of the
defenitions, cross-references and variant spellings encountered (i.e. sulfur,
sulphur: soybeans, soyabeans, Glycine Max L.).  Variant spellings add to the
problem of finding what you want because your full text might have the search
term on way whereas the you are searching for it with a different spelling.  By
having a browsable index you reduce this problem.

Avi Rappoport wrote:

> I do not recommend searching on just keywords.  While it does cut
> down the amount of junk you get back, it also makes it impossible to
> search for vocabulary and concepts that are slightly different from
> those of the indexer.  There is nothing more frustrating than
> searching for an item you *know* is there but can't get to.
>
> A free text search engine that allows users to specify when they want
> All words in a search and a good relevance ranking algorithm should
> give you reasonably good results.
>
> Check out my site, <http://www.searchtools.com/> for information on
> all available site, Intranet and portal search engines.
>
> Best of luck,
>
> Avi
>
> At 9:41 AM -0700 9/24/99, Luck, Deanne wrote:
> >We are in the midst of a total redesign of our web site, and I want to
> >include a "search this site" feature, which we've never had before.  I often
> >find searches on other sites to be very unhelpful due to either the display
> >of the results or the fact that it's searching the full text and returns too
> >much.  I am looking into assigning metadata keywords to our pages, and by
> >default only searching those keywords.  My questions for anyone who has done
> >this or has thought about it are:  Is it worth the trouble to assign the
> >keywords, or is it almost as useful just to have descriptive titles?  If so,
> >how many keywords should be assigned to each page?  Should we develop a
> >controlled vocabulary?
> >
> >Also, what are the best packages to use for both search options and a good
> >results display?  Is SWISH-E far and away the best?  We are running IIS on
> >an NT machine.
>
> ________________________________________________________________
> Avi Rappoport, Search Tools Maven: <mailto:avirr at lanminds.com>
> Guide to Site Indexing and Local Search Engines: <http://www.searchtools.com>



--
Thomas Edelblute
Anaheim Public Library