Sisyphus III

1
Sisyphus III

• Shari Holstege
• 15 November 2001

2
Topics

• The Myth of Sisyphus
• Sisyphus Problems
• Modeling real data
• Formal Concept Analysis

3
The Myth of Sisyphus

• Sisyphus was the king of Corinth, or Ephyra. He
  tricked Hades, the king of the Underworld,
  several times. As his punishment, he must roll a
  tremendous boulder up to the top of a hill over
  and over again, since it rolls back down every
  time he reaches the summit.

4
Why Sisyphus Problems?

• Early on, developers tried to share knowledge
  acquisition, modeling, and management tools to
  evaluate them, but these attempts often failed
  due to differences in platform and support.
• As an alternative, the Sisyphus challenge
  problems were developed. A knowledge acquisition
  problem was defined, and developers are
  challenged to solve it with their own tools.

5
Overview of the Sisyphus Problems

1. Room Allocation
2. Elevator Configuration
3. Igneous Rock Classification
4. Integration over the Web

6
Sisyphus III Objectives

1. To provide for better quantitative comparison of
   KB systems and the methodologies employed to
   build them, through use of a set of achievement
   metrics
2. To provide more realistic access to actual KA
   material in a staged series of releases
3. To obtain more complete records (or knowledge
   engineering meta-protocols) concerning the
   processes that the knowledge engineer goes
   through in the KBS construction process

7
Sisyphus III

• The project was meant to be a decision support
  system to assist astronauts to classify rocks and
  develop a system to tutor them.
• The data given the developers was real data from
  experts and was not familiar to the developers.
  Their task was to model the data without any
  additional information.

8
Problems with Real Data

• Different experts have different levels of
  expertise
• Different experts use different terminologies
• Transcripts of interviews may be imperfect
• Some information is missing
• Information my be conflicting

9
Formal Concept Analysis (FCA)

• A mathematical way to find, structure, and
  display relationships between concepts
• A formal context is a triple (G, M, I)
• G is a set of objects
• M is a set of attributes
• I is relation between objects and attributes

10
An Example - Beverages
Non-alcoholic Hot Alcoholic Caffeine Sparkling
Tea X X
Coffee X X X
Mineral Water X X
Wine X
Beer X X
Cola X X X
Champagne X X
11
Line Diagram
12
Preparing the Data Resources

• Plain scaling multivalued attributes are
  replaced by several single-valued attributes At
  most one of those attributes can be checked in
  the table
• Example grain size might be divided into
  coarse, not coarse, and fine, and each of those
  gets its own column in the table or dot on the
  line diagram

13
Grain Size and Emplacement
Oops why is rhyolite all by itself?
14
Grain Size and Formation Environment
Now trachyte is the odd one out.
15
Grain Size and Color
Grain size and color are independent.
16
Object Model
17
Converting to the Model

• Resolve possible conflicts revealed by the
  diagram.
• If several attributes hold for an identical set
  of objects, the context can be reduced by
  deleting all but one of the attributes.

18
Conclusion

• Formal concept analysis is a way to take real
  data and break it down into a usable format.
• It was applied to the Sisyphus-III domain of
  igneous rock classification to help find noise in
  the data and to help researchers with little
  domain knowledge model the data effectively.