COM362 Knowledge Engineering

1
Knowledge Modelling and Representation

• John MacIntyre
• 0191 515 3778
• john.macintyre_at_sunderland.ac.uk

2
What is a Knowledge Model?

• Knowledge Engineer must be able to build a map
  or sketch of the knowledge domain
• Varying degrees of formality
• small problems can be informally represented
• larger problems need formal approaches
• Formal approaches to modelling are complex and
  difficult to implement
• Need to understand the theory!

3
Why Model Knowledge?

• Aids the visualisation of the knowledge domain,
  helps KE to understand how components fit
  together
• Helps to identify areas where knowledge
  acquisition is required
• Helps to identify where holes exist in acquired
  knowledge
• Helps to determine best way to represent
  knowledge in the knowledge base

4
Simple Models

• For simple problems, simple models are best
• Over-complication to be avoided
• Can take the form of
• sketches
• structure charts
• tree diagrams
• Need to be consistent in terminology throughout

5
Simple model
6
Representing Knowledge

• Necessary to determine the best way to represent
  knowledge
• Therefore must first characterise the knowledge
• Can then choose from a number of representation
  techniques
• Representation technique should fit knowledge
  model

7
Characterising Knowledge

• Must decompose the problem into manageable parts
• Knowledge about the domain
• Knowledge about the reasoning process and how it
  is controlled
• Knowledge about the interface between system and
  user
• Knowledge about the condition of the knowledge

8
Knowledge about the Domain

• Objects,items or entities in the domain
• repertory grids
• Properties and behaviour of objects
• is there a procedural nature to their behaviour?
• Relationships between objects
• clustered, totally unrelated

9
Knowledge about the Reasoning Process

• Want to know how the reasoning process is
  controlled
• Can we mimic this in the KBS?
• Types of control
• Diagnostic - backward chaining?
• Determine all possible solutions - forward
  chaining?
• Dynamic, on-line - blackboard architecture?

10
Knowledge about theUser Interface

• Need to know if the interface requires
• Explanation facilities
• explain terms and concepts
• Justification for diagnosis/advice/outcomes
• to validate and give confidence
• Checkpoints
• to break up long sessions
• Knowledge sharing
• for multiple users
• Security levels

11
Knowledge about the Condition of the Knowledge

• Dependent upon the knowledge domain
• Is the solution process well defined?
• Is there no obvious control strategy?
• Is the problem solving strategy heuristic?
• Is it possible to extract information which can
  be written as rules?

12
Complex Models

• Simple models will not work in complex knowledge
  domains - true??
• Complex problems will naturally lead to complex
  models
• Various theories and techniques exist
• Need to try to fit together in some sensible way

13
Knowledge Representation Languages

• Two specific forms of KRLs are
• Knowledge Level Representation Languages
• used to express Knowledge Level Models
• Ontology Representation Languages
• used to express Ontologies (examples are
  Ontolingua, CML in KADS)

14
Knowledge Level Models

• A conceptual model of knowledge which ignores
  implementation issues
• Expresses some portion of the knowledge needed to
  achieve problem solving ability
• Typically is constructed in the early stages of
  KBS development
• Specifies requirements for the design and
  implementation of the final KBS

15
K.L. Models are used to...

• Specify requirements of one or more KBSs
• Drive model-based knowledge acquisition
• Evaluate the correctness and/or completeness of
  knowledge
• Improve communication
• Document knowledge in an unambiguous manner

16
Ontologies

• A relatively new concept in KE
• An ontology defines
• the vocabulary (set of terms used for modelling)
• the structure of statements in the model
• the semantic interpretation of terms

17
Structure of an Ontology

• May contain three structures
• Terminological Ontology
• (eg Vocabulary or Term Database)
• Information Ontology
• (eg Conceptual Database Schema or Data Model)
• Knowledge Modelling Ontology
• (eg Knowledge Base Structure)

18
Other Ontologies

• Other important types of Ontology are
• Domain Ontology
• (contains specific information about the problem
  domain)
• Problem Solving Method/Task Ontology
• (dedicated to a specific task, solving a
  particular type of problem)
• Generic Ontology
• (concepts generic to several domains)

19
Ontologies in Use

• Medical ontology developed at the University of
  Twente
• Electrical network ontology developed by Labein,
  Spain
• Enterprise ontologies developed by AIAI,
  Edinburgh University and University of Toronto
• Ship design ontology developed by Lloyds Register

20
Where Ontologies Fit
21
Knowledge Sharing

• Useful if new systems could re-use existing
  knowledge bases
• Requires standard representation of knowledge at
  the modelling level
• Standardisation is the key - hence the use of
  ontologies
• Domain Ontologies or Problem Solving Methods can
  be re-used

22
K.L. Models in Detail

• Used to construct conceptual models, including
• knowledge acquisition
• re-use and configuration of re-usable components
• validation and verification
• Act as an operational prototype of the KBS,
  supporting model evaluation, knowledge
  acquisition etc.

23
...continued

• Used to develop a KBS where the K.L. Model acts
  as
• functional specification of the behaviour of the
  KBS, or
• a specification for the design of the KBS, and in
  which
• the conceptual model is preserved in the system
  design, or
• the model is automatically transformed into an
  application

24
...continued

• Models can be used for analysis and assessment of
  KBS behaviour
• Support knowledge management decisions
• Can help understanding of KBS behaviour for
  potential users and developers
• Models can be used to help in the integration of
  systems
• Preservation of knowledge

25
Re-Use of KL Models
26
K.L. Formalism Attributes

• Expressive power
• Scalability and complexity
• Support for knowledge acquisition
• Support for derivation from existing structures
• Support for gradual refinement
• Support for validation
• Support for verification

27
More Attributes

• Support for execution
• Support for mapping
• Support for models with
• Modularity
• Layering
• Distinguishing categories of knowledge
• Distinguishing static/dynamic control
• Separation of generic and application-specific
  parts

28
More Attributes

• Degree of formality
• Representational form
• Computational paradigm
• Guidance on methodology
• Relationship to standards
• Use by other communities (eg mainstream software
  engineering)

29
Formal Approaches to Modelling

• CommonKADS - CML/FML
• DESIRE
• EXPRESS/STEP
• FORKADS
• KADS
• KARL
• KL-ONE
• MODEL-K

30
Problem Solving Model Attributes

• Support of software re-use
• Development of similar applications
• Domain independence
• Independence from implementation details
• Library support
• Support for knowledge acquisition
• Support for systems development
• Automatic compilation
• Integratability

31
...continued

• Support to IPS life-cycle
• Explanations/documentation
• Modifications
• Software re-engineering
• IPS quality support
• Validity and completeness
• Efficiency and transparency

32
Representing Knowledge

• Model development and knowledge characterisation
  should indicate how to represent knowledge
• Many different techniques
• Rules
• Frames
• Semantic Nets
• Logic and Sets
• Blackboards

33
Representing Knowledge

• Keys to the choice of representation technique
  come from knowledge characterisation
• Is the knowledge best represented by objects?
• Is it possible to express the knowledge as rules?
• Is the problem solving strategy dynamic?
• Is it easy to associate particular words or
  phrases with parts of the knowledge domain?

34
Formal Knowledge Representation

• Knowledge-based systems use a Knowledge
  Representation Language (KRL)
• An artificial language which allows a symbolic
  representation of knowledge
• Two main uses of KRLs
• Aid in conceptualising a model of the knowledge
• Construction of a knowledge base

35
K.L. Models - Key Points

• K.L. Models can be categorised into
• Model Components (re-usable)
• Models (non re-usable)
• Examples of each category are shown
• Re-usable components used to develop non
  re-usable models
• The same KRLs can be used to express the
  re-usable and non re-usable components

36
Conclusions?

• MUST build a knowledge model before trying to
  build a system!
• Complexity of model should reflect complexity of
  the problem and the domain
• Model should be used to select appropriate
  representation technique
• Model should help to drive knowledge acquisition