Knowledge Acquisition

1
Knowledge Acquisition
see Course Readings Extracting expertise from
Experts Methods for Knowledge Acquisition p. 35
2
Knowledge Acquisition

• Knowledge is elicited from the expert (or other
  sources), represented and then coded into the
  knowledge base.
• elicitation techniques
• manual
• semi-automated
• automated
• representation schemes could be
• decision trees
• influence diagrams
• rule matrices
• conceptual graphs

3
Elicitation techniques manual approaches

• unstructured/structured interviews  
• good for identifying facts and procedures in the
  problem domain.
• can gain confidence, normally used at the
  beginning of the acquisition phase.
• documented by transcripts leading to a graphical
  representation using a technique such as
  conceptual graph/influence diagram

4
Elicitation techniques manual approaches

• protocol analysis  
• the expert performs the task and verbalises the
  decision making (documented by transcripts).
• Several strategies are possible
• perform familiar or common tasks, then make it a
  bit harder
• perform familiar tasks but withhold some
  information
• perform familiar tasks under a time constraint
• perform difficult tasks by using tough data

5
Elicitation techniques manual approaches

• Advantages and disadvantages of protocol analysis
  
• expert consciously considers decision making
  heuristics
• but this requires an awareness on the part of the
  expert
• expert consciously considers decision
  alternatives,  attributes and values
• but this requires that the expert can
  categorize major decision alternatives
• developer can observe and analyze decision making
  behaviour
• but this requires the expert is able to verbalize
  the attributes and values of a decision variable

6
Elicitation techniques manual approaches

• Advantages and disadvantages of protocol analysis
  
• and finally
• the developer can later check the decision making
  behaviour with the expert
• but is this all a subjective view. Does the
  explanation match the actual reasoning? or
  does it matter

7
Elicitation techniques semi-automated approaches

• These approaches are sometimes referred to as
  indirect methods
• The expert is not being asked to express their
  knowledge directly, instead they carry out
  tasks' and from the results a structure for the
  problem domain can be determined.
• The methods themselves are taken from Cognitive
  Psychology.

8
Elicitation techniques semi-automated approaches

• 1. Multidimensional Scaling  
•   Trying to establish similarity between
  concepts/objects  across a number of dimensions.
•   The expert estimates similarity b/w each
  possible pair  on some scale (0 - 1, 0 9, 1
  -16 etc)
•   Analysis of similarity matrices leads to the
  concepts/objects being clustered in space
• The expert can inspect and describe this
  diagram.

9
Elicitation techniques semi-automated approaches

• 1. Multidimensional Scaling
• example
• the concepts Farm domestic animals goat,
  cow, sheep, pig, horse, dog, rabbit
• dimension size i.e. how similar are they in
  terms of size?
• A similarity matrix can be constructed

10
Elicitation techniques semi-automated approaches

• The higher the number the less similar the
  concepts are

11
Elicitation techniques semi-automated approaches

• Many more similarity matrices could be formed
  using different dimensions e.g. behaviour,
  temperment, suitability for human consumption etc
• A clustering algorithm is then applied to the
  matrices to produce a diagram that shows how
  similar (in an overall sense) the concepts are.
• See figures 7, 8 9 on pages 41-42 of the course
  readings

12
Elicitation techniques semi-automated approaches

• 2. Repertory grid analysis
• The focus is now on difference as well as
  similarity
• from clinical psychology - personal construct
  theory
• important objects/concepts are identified
• The concepts are represented to the
  client(expert) in groups of three
• The expert is asked which one is different from
  the other two? and then in what way? (the
  answer to the 2nd question provides a construct)
• The expert is then asked to rate the other
  concepts in terms of the construct
• (on some lickert style scale)

13
Elicitation techniques semi-automated approaches

• 2. Repertory grid analysis
• This process continues until a series of
  constructs are built up. A rating grid can then
  be produced (e.g. see fig 19 p. 47)
• From this a distance matrix and a hierarchy chart
  can be developed (see figures 20 21 p.47)

14
Elicitation techniques Summary

• Activity identifying general decision making
  heuristics and concepts.
• Technique structured interviews.
• Activity identifying routine tasks or
  procedures during the decision making.
• Technique structured interviews, protocol
  analysis.
• Activity identifying major concepts,
  relationships i.e a semantic view of the problem
  domain.
• Technique semi-automated approaches such as
  multi-dimensional scaling, repertory grid
  analysis.