Semantic Web Research: Ontology Engineering

1
Semantic Web Research Ontology Engineering

• Aditya Kalyanpur
• University of Maryland

2
Outline

• Previous Work
• Ontology-based Annotation SMORE
• Software Engineering Mapping OWL to Java
• Current Research
• SWOOP OWL Ontology Editor
• Browsing, Editing, Search, Annotation, Change
  Management
• Debugging Ontologies (PhD Thesis)
• Glass and Black Box Debugging Services
• Future Plans
• SWOOP next stepsversion control etc.
• Repairing OWL Ontologies
• Explanation of Subsumption,,Query Answering
• Modularizing OWL Ontologies..

3
Previous Work
4
Ontology-based Annotation

• SMORE
• Tool to markup HTML documents with Web Ontologies
  (aimed at average web users)
• Features
• Annotate Text / Photos etc.
• Triple-based Data Generation
• Auto-inferencing of triples based on semantics
  defined in Ontology
• Multiple Ontology Support
• Search/Link NL phrases with ontology terms
• Constraint checking identify invalid triples
• Current plans to extend it to support blogging

5
SMORE
6
Ontology Engg. ltgt Software Engg.

• Mapping OWL to Java
• Aim
• Use OWL as a specification for Java Applications
  (analogous to UML)
• Build applications that mirror existing OWL
  Ontologies
• Desirable Features of Mappings
• Preserve formal semantics as much as possible
• Ensure interoperability of resultant OO code
• Easily tweakable
• Result
• A concise and elegant solution to map an OWL
  Ontology into Java
• Covers most of OWL-Full expressivity
• IBM Development Team built an API based on
  mappings JASTOR

7
Current Research
8
Ontology Editing

• Swoop Web-browser-like Ontology Editor
• Browsing / Editing / Reasoning with Multiple OWL
  Ontologies
• Based on the WonderWeb OWL API (shares some
  design goals)
• Recent release v2.2 has 5000 downloads
• Over 200 Users on Swoop mailing lists ( still
  growing!)
• Numerous Publications on its Use and Extensions
• A place for experimenting our research ideas

9
SWOOP

• Key Features
• Hypermedia-inspired Editor
• Designed for OWL Rec.
• Species Validation
• Presentation Syntax Tabs
• Multiple Local/Web Ontology Support
• Collaborative Annotation (using Annotea)
• Ontology Debugging (using Pellet)
• Automatic Partitioning of Ontologies (based on
  E-connections)
• Resource Holder for comparison/mapping

10
SWOOP
11
Ontology Debugging

• Problem Debugging defects in an OWL Ontology can
  be hard since OWL (DL) is based on an expressive
  description logic,
• Newcomers to OWL have difficulty in understanding
  and fixing defects
• Even DL experts find it hard to debug errors in
  large, complex ontologies
• Main Drawback leads to weakening ontology
  (under-specifying)
• Solution (same spirit as software debugging) an
  OWL Debugging Tool which,
• Explains causes of contradiction
• Pinpoints problematic components
• Suggests remedies

12
Debugging Services

• Explaining cause of error in a single
  unsatisfiable class
• Finding minimal set of axioms in ontology
  responsible for error Sets of Support (SOS)
• Finding the root clash
• Revealing dependencies between unsatisfiable
  classes
• Identifying Root / Derived Unsatisfiable Classes

13
Two Approaches to Debugging

• Glass Box (Reasoner Dependent)
• Modify internals of the reasoner that is used to
  detect the defects
• Need to reduce overhead in normal reasoning
  performance
• Currently used to find root clash and SOS
• Black Box (Reasoner Independent)
• Use reasoner as an Oracle and perform separate
  analysis on the side
• Need to minimize reasoner calls
• Currently used to find Root/Derived unsatisfiable
  classes using structural analysis

14
Glass Box Debugging for SOS

• Example to capture Sets of Support
• MeatPizza v Pizza u 9hasTopping.Meat
• range(hasTopping, Food)
• Pizza v 8hasTopping.Cheese
• Meat v Food u Vegetable
• Cheese v Meat

MeatPizza, (Pizza u 9hasTopping.Meat)1
MeatPizza
MeatPizza, Pizza1, 9hasTopping.Meat1
MeatPizza, Pizza1, 9hasTopping.Meat1,
8hasTopping.Cheese1,3
hasTopping1
Clash! SOS 1 1,3,5 1,3,5
Meat1
Meat1, Food1,2, Vegetable1,4
Meat1, Food1,2, Vegetable1,4, Cheese1,3,
Meat1,3,5
Meat1, Food1,2, Vegetable1,4, Cheese1,3
Meat1, Food1,2
15
Glass Box Debugging for SOS

• Why is axiom tracking hard?
• As expressivity increases, tableaux operations
  become complex
• e.g. Node merges due to cardinality restrictions
• C v 9 R.D u 9 R.E u 1.R
• Internal modifications done by the reasoner
• Normalization involves rewriting terms
• (9 P.C) 8 P. C
• Absorption - involves combining axioms
• (i) C u D v E
• (ii) C v F

C v F u (E t D)
16
Glass Box Debugging for SOS

• Glass Box Evaluation Reasoning Time / Memory

17
SOS Use Case

• SWEET-JPL (1537 classes, 2250 axioms) 1 unsat.
  class

Description of clash
Minimal set of axioms from ontology responsible
for error (clash)
18
SOS Drawbacks

• For a large no. of unsatisfiable classes
• Going down the list and fixing errors one by one
  is time-consuming, pointless and sometimes
  harmful
• Need to identify critical errors (roots) that
  propagate and cause errors elsewhere (derived)
• Examples of propagation of errors
• A v B, where B is unsatisfiable
• A v 9 R.B, where B is unsatisfiable
• A v B t C, where both B, C are unsatisfiable

19
Dependency between Unsat. Classes

• Root and Derived
• Root named class whose unsat. does not depend on
  the unsat. of another named class
• E.g. A v B u B
• Derived named class whose unsat. depends on the
  unsat. of another named class
• E.g. A v B u C, where C is unsatisfiable

20
Black Box Debugging for Root/Derived

• Asserted Dependencies (Relatively
  Straightforward)
• Example
• A v B t C, and both B,C are unsatisfiable
• Detection Technique Structural Tracing
• Inferred Dependencies (Challenging)
• Example
• A 9 P. D
• B 9 P. E u C u C
• D v E
• Detection Technique No known complete black-box
  algorithm
• Heuristics Developed Weak Removal,
  Subsumption-safe Transformations

21
Root/Derived Use Case

• For Tambis (395 Classes, 590 axioms) 144 unsat.
  Classes
• Two Iterations to debug completely

Root Unsatisfiable Classes in the Ontology
Dependency Table
22
User Study

• How effective is SOS/Root-Derived
• University 8 unsat. classes (5 roots/ 3
  derived)
• Sweet-JPL 1 unsat classes
• Mini-Tambis 30 unsat classes (5 roots/ 25
  derived)

Pilot n12 subjects
23
Future Plans
24
SWOOPnext steps

• Expressive TBox Query
• Find all classes ?C such that properties name,
  age, mail have ?C as its domain?
• Find all classes ?C that contain 9 hasChild.Male
  in its definition?
• Version Control system for Ontologies
• Annotea client server base
• Annotate ontology versions directly (metadata
  about change)
• Store updates only, starting with a base ontology
  (easier to obtain diff)
• Difficult to characterize version-clash

25
Debuggingnext steps

• Semi-automated techniques for Ontology Repair
• Identify common conflicts among roots
• Rank axioms (e.g. based on minimal effect causing
  changes)
• Suggest reasonable edits
• Use of Tableaux Tracing
• Explanation of Subsumption..
• Detection of Redundancy

26
Modularizing OWL Ontologies

• Bernardos dissertation on methods for combining
  and integrating ontologies on the Semantic Web
• Theoretical Framework to modularize OWL
  Ontologies based on E-connections
• Automatic Partitioning algorithm based on theory
• Implementation in SWOOP
• Next Steps
• Change management given modules, i.e., a
  framework for collaborative evolution

27
Questions?

• Thanks

28

• Extra Slides

29
Clashes

• Basic Types of Clash
• Atomic Individual belongs to OWL Class and its
  complement
• E.g. C v D u D
• Cardinality Individual has maxCardinality
  restriction but is related to more Individuals
• E.g. C v 1. R u 2. R
• Datatype Literal value violates range
  restriction on OWL Datatype Property
• E.g. C v 9 R. false, where R is a OWLDatatype
  Property with range xsdinteger

Back
30
Structural Tracing

• Root/Derived Detection Examples
• A v C u D (D is unsat.)
• A v (C t D) (Both C, D unsat)
• A v 9 R. C (C is unsat)
• A v 9 R.C (domain(R) unsat)
• A v 1. R u 8 R. C (C is unsat)
• A v 9 R-. C (range(R) is unsat.)
• A B , B v C u C (Inferred)

Back
31
Inferred Dependency (1)

• Weak Removal Approach (heuristic)
• Basic Idea Given unsat. classes A, B, remove
  A, and test satisfiability of B
• Remove tricky to define (motivation make A
  satisfiable, but not alter KB significantly)
• One way Remove primitive definitions of A (Weak)
• Based on two assumptions about Weak-Removal(A)
• A becomes satisfiable
• Parents(A) do not become satisfiable

Back
32
Inferred Dependency (2)

• Subsumption-safe transformations
• Approximate KB by getting rid of unsatisfiability
  while preserving subsumption
• e.g. A 9 P. D
• B 9 P. E u C u C
• D v E
• Replace C by new C, then discover A v B
• Sound but not complete w.r.t subsumption relation
  revealed

Back