[912] in Public-Access_Computer_Systems_Forum
Re: CLR-OPAC study
daemon@ATHENA.MIT.EDU (Charles Hildreth)
Tue Aug 4 11:03:58 1992
Date: Tue, 4 Aug 1992 09:56:11 CDT
Reply-To: Public-Access Computer Systems Forum <PACS-L%UHUPVM1.BITNET@ricevm1.rice.edu>
From: Charles Hildreth <hildreth@eagle.sangamon.edu>
To: Multiple recipients of list PACS-L <PACS-L%UHUPVM1.BITNET@ricevm1.rice.edu>
----------------------------Original message----------------------------
>
> > Charles Hildreth,
> >
> > I have similar interest as you do, but I have no resources to do
> > an extensive study. There is a commercial product called
> > KNOWLEDGE FINDER by Aries Systems. It is a CD-ROM product for
> > MEDLINE. It uses probablistic model for retrieval, so user can
> > enter the query in natural language and the software does the
> > retrieval automatically. I did a simple comparison between the
> > result with the result that I got from a boolean search. Not
> > very good!
> >
> > Kathy Kwan
> > Systems Librarian
> > Winthrop-University Hospital
> > kwan@winthrop.org
> >
Kathy, and PACS-Lers:
I can't tell from this which results set you judged to be the best, the
Boolean search or the "automatic" probabilistic search. But this gives me
a wonderful opportunity to step up on my soapbox.
Boolean exact match searches do not rank the documents retrieved, and do not
retrieve at all documents that partially match the user's query, and may also
be relevant.
Any probabilistic retrieval system worth its "weight" (pardon the pun) will
not only retrieve all the citations/documents a Boolean search would retrieve
for a given query (search statement) but might well retrieve additional relevant
citations/documents and would rank all of them for the user in decreasing order
of probable relevance to the query. To repeat, none of the Boolean results
would be "unretrieved" though some of them might be lower down the relevancy
list.
A good probabilistic retrieval system (vendors can screw up anything) would
invite relevance feedback from the searcher regarding these results, and
would continue the search and/or rerank the citations/documents in the
search results set.
Regarding today's easy-to-use, sneaky OPACs, would you rather the system
put a Boolean AND or a proximity WITH between your search terms (they do
not tell you which), or would you have the system retrieve all citations/
documents that match your query terms in any degree and present them to you
in an intelligently arrived at order of relevance to your search query?
Some retrieval system researchers have called probabilistic systems "partial
match" retrieval systems. They are better labelled as "best match" or "full
match" systems, and they work harder for the user than Boolean exact match
systems.
Show me an end-user who likes Boolean operators and I will show you ....
Regards, Charles Hildreth (hildreth@eagle.sangamon.edu)
