KBS Development on the Semantic Web

1
KBS Development on the (Semantic) Web

• David Corsar
• Derek Sleeman
• Department of Computing Science
• University of Aberdeen

2
Overview

• Introduction
• Building Knowledge Based Systems from components
• Related Work
• Our Approach
• Example
• Applying Approach on the (Semantic) Web
• Summary

3
Knowledge Based System (KBS)

• A KBS is
• An A.I. System
• Which uses domain knowledge
• Combined with some intelligent reasoning
• To solve typically a nondeterministic task which
  requires significant human resources

4
Building KBSs

• Building everything from scratch every time
  proved to be very expensive
• Cheaper to build by reusing existing components
• Domain Knowledge (instantiated ontologies)
• Generic Reasoning Algorithms (Problem Solving
  Methods (PSMs))
• Configured to work together as required

5
Challenge

• Various projects looked at building KBSs by using
  reusable components from repositories
• These projects produced good theoretical results
• However, for various reasons, dream still to be
  fully realised
• We aimed to develop a methodology and
  implementing tool which aids the user in creating
  new executable KBSs from components

6
Related Research - CommonKADS

• Major European project
• Complete KBS development methodology (c.f.
  software engineering)
• Construction of various models of organisation (6
  in total)
• Expertise model descriptions of 11 PSMs
• Little supporting implementation developer uses
  descriptions as a guide for implementing domain
  model and reasoning algorithm

7
Related Work - PSM Librarian

• KBS development by
• Selecting a PSM/method ontology from a repository
• Selecting a domain ontology
• Mapping domain ontology to method ontology, using
  mapping ontology
• Implemented as PSMTab for Protégé
• No specific support for defining mappings
• No support for executing the configured KBS

8
Shortcomings

• These projects made several significant
  theoretical contributions
• However, implementations were often inadequate
• Each developed different formalisms for defining
  ontologies, PSMs, etc
• Based on false believe that domain ontology
  from one KBS would meet the requirements of a new
  KBS
• Domain ontology developed for one particular
  application contains procedural knowledge
  required by that application, and so is unlikely
  to meet the requirements of a different
  application, which will require different
  procedural knowledge

9
Technical Developments

• OWL
• SWRL, RIF
• Protégé
• Mature environment for building, editing,
  instantiating and using ontologies
• JessTab plug-in for reasoning with instantiated
  ontologies
• Semantic Web (Berners-Lee et al 2001)
• Natural language text conveys knowledge to humans
• Corresponding (instantiated) ontologies provide a
  form of easily accessible knowledge
• Ontologies associated with mapping rules

Mapping Rules
10
Our Approach

• Two stage methodology for building KBSs through
  reuse
• Perform automatically as much as possible, while
  supporting user when he/she needs to make
  decisions/selections
• Ontology mapping for reuse of domain knowledge
• Directed KA to acquire the rules which are
  required by a generic PS to reason in domain

11
Illustrative Example

• Elevator Diagnosis
• Elevator components
• Diagnostic algorithm
• Faults
• Symptoms
• Repairs

?

• Elevator Configuration
• Elevator components
• Propose and Revise
• Configuration Rules
• Constraints
• Fixes

12
Methodology Initial Components
KBS(diag, elevator) PS(diag, elevator)
ONT(elevator, diag)
ONT(elevator, diag)

• Select the generic propose and revise Problem
  Solver (PS)
• Composed of
• Ontology describing PS rules
• Some generic PS code

• Extract the elevator domain ontology from
  existing KBS

13
Problem Solvers

• A generic problem solver is
• A domain independent description of a particular
  type of reasoning (e.g. diagnosis, propose and
  revise)
• Provided by an ontology, PS-ONT which describes
• The rules the PS uses to work in a domain
• The form of the domain knowledge it requires
• Also contains, an implementation of any generic
  PS code (PS-RS)

14
Methodology Stage 1Mapping
ONT(elevator, diag)
KBS(diag, elevator) PS(diag, elevator)
ONT(elevator, diag)
Mapping
15
Ontology Mapping

• The generic PS requires domain knowledge
• To provide the context in which domain rules are
  acquired in Stage 2
• To provide domain knowledge for the final KBS to
  reason with
• User assisted in defining mappings between the
  domain ontology ONT(elevator, diag) and the PS
  ontology PS-ONT(pnr, -)
• Mappings executed adds domain knowledge to PS
  ontology to give PS-ONT(pnr, elevator)

16
Methodology Stage 1Mapping
ONT(elevator, diag)
KBS(diag, elevator) PS(diag, elevator)
ONT(elevator, diag)
Mapping
PS-ONT(pnr, elevator)

• After mapping is completed we have
• The PS ontology
• Enhanced with (elevator) domain knowledge

17
Methodology Stage 2Focused Knowledge
Acquisition
ONT(elevator, diag)
KBS(diag, elevator) PS(diag elevator)
ONT(elevator, diag)
Mapping
PS-ONT(pnr, elevator)
Knowledge Acquisition
18
Methodology Stage 2Focused Knowledge
Acquisition

• PS-ONT describes the structure of the rules that
  the PS uses to reason in a domain
• Rule descriptions based on SWRL
• Rule description consists of
• Rule name
• The valid types of antecedents and consequents
• How rules are related e.g. pnr a constraint
  rule is associate with fix rule

19
Example Interaction Defining new Constraint
Fix Rules
Constraint

• SYS What are values that define a violation
  related to total-cab-weight?
• USER total-cab-weight gt motor-supported-weight
• SYS What is the name of this violation?
• USER Cab too heavy for motor
• SYS What actions should be performed if this
  constraint violation is present?
• USER Replace the motor with one thats
  supported-weight gt total-cab-weight
• paraphrased version of the user/system
  interaction

Fix
20
Methodology Stage 2Focused Knowledge
Acquisition
ONT(elevator, diag)
KBS(diag, elevator) PS(diag, elevator)
ONT(elevator, diag)
Mapping
PS-ONT(pnr, elevator)
PS-ONT(pnr, elevator)
Knowledge Acquisition
21
Methodology Generating Executable KBS
ONT(elevator, diag)
KBS(diag, elevator) PS(diag, elevator)
ONT(elevator, diag)
Mapping
PS-ONT(pnr, elevator)
PS-ONT(pnr, elevator)
Knowledge Acquisition
Rule Generation
PS-RS(pnr, elevator)
22
Implementation

• MAKTab a Protégé plug-in
• Generic PSs for diagnosis and propose and revise
• Currently have rule generators which produce
  executable JessTab KBSs
• System has been used to build elevator diagnosis
  and pnr KBS by developer and by users to build
  computer hardware diagnosis and pnr KBS

23
KBS Development on the Semantic Web

• The Semantic Web will potentially
• Provide access to many more ontologies
• Provide details of how to map between these
  ontologies
• These can be used to provide an improved
  implementation of our approach
• Developing a Web based system should also make
  KBS development (and execution) more accessible

24
Related Work

• IBROW3
• Intelligent broker for KBS development on the Web
• Prototype broker developed
• No longer available today
• Internet Reasoning Service 3
• Semantic broker for developing applications of
  Semantic Web Services

25
Considerations

• Finding components
• Domain ontologies
• Problem solvers
• Storing PSs
• Mapping
• Generating and executing KBSs

26
Domain ontologies

• Finding
• Ontologies should be much more prevalent on the
  Semantic Web than they are today
• Ontologies from Semantic Web sites could be used
• Ontology search engines and repositories
  (ONTOSEARCH2) could also be used
• Evaluating and fixing ontologies
• CleOn and RepairTab could be used to improve the
  quality of an ontology before it is used in KBS
  development

27
Online PSs and KBS

• MAKTab
• PS accessed by loading PS ontology
• Easy for developers to add new PSs associated
  rule generators MAKTab automatically uses these
  to generate KBS
• User runs KBS by copy and pasting into JessTab
  (which is bundled with Protégé)

• Online
• PS selected from a centralised collection
• Online system could provide open PS repository
  and generate KBS, but must then trust contributed
  rule generators
• Can not rely on user having any specialised
  software online system must run KBS

28
PS on Web - Possibilities

• Central Repository
• Stores all generic PSs
• Responsible for generating and running KBSs
• Enforces developers to generate KBSs in
  predetermined languages

• Registry
• PSs as Web Services
• Maintain registry of PS WSs
• WS responsible for providing ontology, generating
  and executing KBS
• Allows developers to choose KBS execution
  language

Tool Interface
Tool Interface
PS1
PS Repository

PS Registry
PS1
PSn

PSn
29
Mapping

• Ontologies on Semantic Web are associated with
  mapping rules
• Online system will store defined mappings, making
  it possible to build a repository of mappings
• Both of these can be used to potentially improve
  the suggestion of mappings provided to the user

30
Supporting Mapping - 1
Associated Mappings
31
Supporting Mapping - 2
Mapping Repository
32
Supporting Mapping - 3
BuildingParts.com
Human Pages
Parts Ontology
2
1
Mapping Repository
33
Supporting Mapping - 3
BuildingParts.com
Human Pages
n-1 Ontologies
Parts Ontology
.
2
n-1
1
Semantic Web Site n
Human Pages
Ontology n
Mapping Repository
n
34
Summary

• We have developed a methodology and supporting
  tool for KBS development
• Propose a (Semantic) Web based implementation of
  methodology that supports KBS development using
  (reusable) components available on (Semantic) Web
• Semantic Web should provide wealth of new
  components that can be used for KBS development
• Potential to further automate and improve our
  tool for KBS development, making it more
  accessible

35
Acknowledgements

• The Protégé team
• Mark Musen and colleagues for making available
  their version of the Sisyphus-VT code
• Henrik Eriksson (JessTab)
• MAKTab work was supported by the Advanced
  Knowledge Technologies (AKT) IRC.