| Some of the work done by AIRG (John McDermott' group) *might* be
classified as automated KA, though I think they prefer different
names, e.g. "situated computing". You may want to contact them;
if their work isn't what you're looking for, they'll probably be
able to tell you what's going on in the field.
There is/was some r & d at C-MU on rule generation; I think Tom
Cooper was tracking it.
NEXPERT and possibly other shells claimed to have some automated KA.
I'm suspicious but don't have any direct experience with them. Does
anyone else?
|
| Malcolm,
A few more references for (semi)automated KA tools & techniques :
- induction languages with rule generation :
. Expert-Ease, Rule-Master, Nextra, KATE
- model-based tools possibly with rule generation :
. MORE : generates OPS5 rules for diagnosis tasks
. SALT : same for configuration
- hypertext-style tools to get an abstract expertise model from
interview reports : K-station (ILOG), coming from KOD
methodology (cf. book of Claude Vogel)
- KADS methodology with Shelley prototype (unfortunately on
SUN's only at this time) : modelling of expertise &
application design (based on problem-type identification).
Hope it helps,
-- St�phane --
|
| People interested in automated KA may want to contact Jude (pronounced
same as "Judy") to get slides from this talk if they're available.
Mark
.............................................................
From: FASDER::SELECT::LMOADM::TSS "JUDE' PARTRIDGE - AITC OPERATIONS, DTN 296-5758, LMO2-1/J3 03-Dec-1991 1804" 3-DEC-1991 18:27:21.00
To: @CC$PUBLIC:MASTER.DIS
CC: TSS
Subj: Paul Compton, Fri 12/13, 10AM, Room 144A, GARVAN-ES1
TITLE: "Ripple Down Rules for Knowledge Acquisition:
Better Than You Think"
SPEAKER: Paul Compton
School of Computer Science and Engineering
University of New South Wales
Kensington, NSW, Australia
DATE: Friday, December 13, 1991
TIME: 10:00 AM - 12 NOON
PLACE: LMO2, Room 144A
HOST: David Marques, Technical Staff Member
AI Research Group, AI Technology Center
NOTE: This presentation will *NOT* be videotaped.
-- -----
A major problem with building expert systems is that experts
always communicate knowledge in a specific context. A knowledge
acquisition method has been developed which restricts the use of
knowledge to the context in which it was provided. This method,
"ripple down rules" allows for extremely rapid and simple knowledge
acquisition, where the time required to incorporate each new piece
of knowledge remains constant, regardless of the size of the knowledge
base.
An expert system (GARVAN-ES1 -- interprets pathology laboratory
reports) based on this approach, and built by experts without the
help of a knowledge engineer, is in routine use.
To capture context, the expert system is built as a tree with a
rule at each node with two branches, depending on whether the rule
is satisfied or not. Any new rule that is added in response to a
wrong interpretation is attached to the branch at which the expert
system terminated, thus making a new node.
We found that no knowledge engineering was required and that
'ripple down rules' could be simply added to the knowledge base
as provided by the expert.
This results in knowledge acquisition at least 40 times as fast
as that required for a conventional version of the same knowledge
base, with the same knowledge engineer/expert involved.
The talk will present some technical discussion of how the knowledge
base is built, as well as data on the performance and growth of the
system in use.
This research was sponsored in part by Digital Equipment Corporation.
|