[674] in Public-Access_Computer_Systems_Forum
Next generation OPAC
daemon@ATHENA.MIT.EDU (Millard Johnson)
Mon Jul 6 09:58:28 1992
Date: Mon, 6 Jul 1992 08:46:18 CDT
Reply-To: Public-Access Computer Systems Forum <PACS-L%UHUPVM1.BITNET@ricevm1.rice.edu>
From: Millard Johnson <johnsonm@ohsu.EDU>
To: Multiple recipients of list PACS-L <PACS-L%UHUPVM1.BITNET@ricevm1.rice.edu>
----------------------------Original message----------------------------
Content-Type: Text
Content-Length: 0
Content-Type: Text
Content-Name: C:\MJ\MAIL\CHUCK6.TXT
Content-Length: 3638
Concerning next generation OPAC design:
I agree that relevance feedback is important, but it should
be focused. *Johnny's big book of atoms * may be fine for
one audience and usless for another. The question is - what
books and papers have been found to be important to
audiences composed of people like me? I may want to turn
off the relevance ranking, but that choice should be mine.
So again, the relevance feedback needs to know something
about the evaluator. (By the way this should probably be
shared information. A national file of relevance ranking -
something like ISI citation counting. That way the authors of
outstanding works can get the kind of credit that the authors
of lots of works now get.)
In the late 70s I spent LOTS of hours analyzing logs of
online catalog usage. Some of the things I found:
1. Physicians and medical students know lots of dirty words.
2. The greatest source of error was the user selecting the
wrong tool from the bag of tools available over the OPAC.
Every OPAC has different tools but they are all designed to
accomplish a relatively few tasks:
A) find a single known item
B) find all works by a single author
C) find all books or journals on a subject
D) browse (plus a few other less frequent things).
Usually there is a single best tool for each objective. Most
OPACS assume the user will know when to use which
technique. This assumption that the user will know which
tool to employ -- is frequently wrong. For example: The best
way to find a single known item is to enter 1 word from the
name field and one or two significant words from the title.
Any other stragegy results in a higher failure rate, and the
rate of failure goes up with every key stroke typed. The
point is that the system should ask the user what he/she
wants to do and then employ the best tool. (The expert
search is another matter. I am talking about the 90% case.)
3. It is not all that difficult to see when the user is failing. It
would be a trivial artifical intelligence task to interrupt the
user when he/she is failing and offer some help.
4. In many instances a failing search can be salvaged by
dropping through a precise search to an imprecise search.
For example -- a miss on a subject term can drop through
to: a subject key word, a title key word, a spelling
approximation. You can tell the user what you did, show
him the results and ask him if he wants to go farther.
5. Subject searching has the highest failure rate because
users do not use controlled vocabulary well. Conceptually,
subject searching is drawing a fence around a multi
dimential subject space. Typing words is an inherently poor
technique for a lot of people. I suggest drawing 4 rectangles
and putting some subject words from different subject
domains in each rectangle. Let the user click somewhere
on the screen. (clicking on the center would get
encyclopedias) Then show four more rectangles of narrower
subject interest. At a lower level you could show the titles of
several books and show the user books like the one he
clicks. Ten or so clicks would, I think, get a lot more people
a lot closer to their subject interest than three times as many
key strokes. (Remember I am talking about the average
undergraduate who is not a trained experienced subject
searcher computer jock.)
****************************************
I would rather risk failure than achieve it without
risk.
Millard Johnson, PORTALS
johnsonm@ohsu.edu
****************************************
