KNOWLEDGE ACQUISITION

1
KNOWLEDGE ACQUISITION

• CONTENTS
• 1. Knowledge engineering
• 2. Methods of knowledge acquisition
• 3. Knowledge Acquisition from Multiple Experts

2

• Knowledge engineering
• Activities
• knowledge acquisition
• knowledge validation
• knowledge representation
• inferencing
• explanation and justification
• Stages of knowledge engineering
• identification
• conceptualisation
• formalisation
• implementation
• testing

3
(No Transcript)
4

• Knowledge elicitation is the process of
  capturing knowledge and presenting it in a way
  that is comprehensible to people.
• Knowledge acquisition is the process of
  transferring and transforming knowledge about
  expert problem solving from the knowledge source
  to a computer program.
• Difference between knowledge-based system
  development and conventional IS development
• characteristics of typical problem domains are
  different
• different architectures
• Sources of knowledge
• documented
• undocumented

5

• Levels of knowledge
• Shallow knowledge
• describes narrow input-output relationship
• example IF gasoline tank is empty THEN car
  will not start
• Deep knowledge
• describes complex interactions and a system's
  operation
• example.

6

• Major categories of knowledge
• declarative
• procedural
• meta-knowledge

7

• Problems in knowledge acquisition
• expressing the knowledge
• transfer to a machine
• number of participants
• structuring the knowledge
• other reasons
• experts may lack time
• testing and refining knowledge is complicated
• knowledge may be scattered across several
  sources
• knowledge collected may be incomplete
• knowledge may be mixed up with irrelevant data
• observation or interviewing may influence
  experts
• communication between the knowledge engineer and
  expert may be problematic

8
Methods of Knowledge Acquisition 1. Manual
methods
Experts
Elicitation
Coding
KnowledgeBase
KnowledgeEngineer
Documented knowledge

• interviewing
• tracking the reasoning process
• observing

9
2. Semiautomatic methods
RepertoryGrid
Experts
KnowledgeBase
Editors
KnowledgeBase
KnowledgeEngineer

• to support experts
• to support knowledge engineer

10
3. Automatic methods
Case historiesand examples
KnowledgeBase
InductionSystem

• induction methods

11

• Interviews
• unstructured
• structured
• Advantages
• explicit knowledge is elicited quickly
• Disadvantages
• difficult to acquire relevant and correct
  knowledge
• lack of overall structure
• time consuming and tedious

12

• Protocol Analysis
• Protocol is a record of the expert's
  step-by-step information processing and decision
  making behaviour
• Advantages
• the knowledge engineer can record and later
  analyse with the expert key decision points
• richness of details
• Disadvantages
• transcripts can be ambiguous
• difficult and time consuming task
• Observations
• Advantages
• most obvious and straightforward approach to
  knowledge acquisition
• Disadvantages
• large quantities of knowledge are being collected
• expensive, time consuming

13

• Expert-Driven Methods

Repertory Grid Analysis 1. Identify the important
objects example. selection of a computer
language objects LISP, PROLOG, C, COBOL 2.
Identify important attributes attributes type,
ease of programming, training time,
availability 3. Trait and opposite for each
attribute type symbolic (3) - numeric (1) ease
of programming high - low training time high -
low availability high - low 4. Assign the
points to each attribute for each object
14

• Advantages
• best suited for well structured problems
  (diagnosis, classification)
• Disadvantages
• no deep knowledge
• no knowledge about decision making behaviour

15
maximum difference 9 8 72 1 0 1 2 1
0 3 0 8
16
5 (C1, C6) 5 (C2, C6) 6 (C1, C4) C4 7 (C4,
C6) 7 (C2, C9)
C6
C9
17
(No Transcript)
18
0 1 1 2 1 0 4 9
19

• Supporting the knowledge engineer
• editors and interfaces
• explanation facility
• revision of the knowledge base

20

• Automated knowledge acquisition - machine
  learning
• Automated Rule induction
• training set ? rules
• Advantages
• suitable for domains which are complex,
  uncertain
• can be used by experts or system analyst
• offer the possibility of deducing new knowledge
• Problems
• may generate rules that are not easy to
  understand
• expert still has to specify the attributes
• good only for rule-based classification problems
• the number of attributes must be small
• require a large number of examples
• exceptions to rules should be removed
• Example. ID3
• Included in ES shells CRYSTAL, K-Vision,
  VP-Expert

21
IF number of joins 1 AND number of curves
0THEN class is A ...
IF number of curves 0THEN class is A IF number
of curves 1 THEN class is B
22

• Knowledge Acquisition from Multiple Experts
• Advantages
• synthesis of expertise
• fewer mistakes
• eliminate the need for using a world-class
  expert
• wider domain then a single expert's
• Problems
• compromising solutions generated by a group with
  conflicting opinions
• difficulties in scheduling the experts
• dominating experts
• Approaches to using multiple experts
• Individual experts
• Primary and secondary experts
• Small groups
• Panels

23
SUMMARY

• Knowledge engineering, knowledge acquisition
  are introduced.
• Methods of knowledge acquisition can be divided
  into manual, semiautomated, and automated.
• The manual methods are interviewing, protocol
  analysis, observing.
• Repertory grid is the most common expert-driven
  semiautomated method.
• Rules induction is the most common automated
  knowledge acquisition method.
• There are benefits as well as limitations and
  problems in using several experts to build a
  knowledge base.

24
References 1. Decision Support Systems and
Intelligent Systems, Fifth Edition E.Turban, Jay
Aronson, Prentice Hall, 1998. 2.
Knowledge-based Decision Support Systems, With
Applications in Business, 2nd Edition, M. Klein,
L. Methlie, Wiley, 1995. 3. Advanced Students'
Guide to Expert Systems, G.Marshall, Heinemann
Newnes, 1990.