Date post: | 27-Mar-2015 |
Category: |
Documents |
Upload: | samantha-ray |
View: | 212 times |
Download: | 0 times |
Re-learning Ontology Management for the Web
Chris Welty
IBM Research
Web-based Ontologies
• The “symbols” in the ontology are URIs
• The “meaning” of the ontology is distributed around the web
• The entire ontology may not always be accessible
The Web enables…
• Searching for, ranking ontologies– But not based on meaning
Searching, Ranking, & Quality
• Does quality matter?• Good quality ontologies cost more
– Coverage, correctness, richness, commitment [Kashyap, 2003]– Organization, meta-level consistency [Guarino & Welty, 2000]
[Rector, 2002]– Required for some applications
• Improvements in quality can improve performance [Welty, et al, 2004]– 18% f-improvement in search– Cleanup cost ~1mw/3000 classes– BUT … low quality ontology still improved base
• But to rank quality, it needs to be measured…
The Web enables…
• Searching for, ranking ontologies– But not based on meaning or quality
• Selection, Reuse of ontologies– Including misuse
• Partitioning ontologies– Reusing parts of an ontology
… A problem looms
AT&T Definity System (c. 1992)
• 20,000,000 lines of C code• 1,000 programmers• High turnover
– 25% less than 1 year experience– 75% less than 5
• High reliability requirement – “1 min/year”– “Handle Mother’s day call volume”
• 6+ dimensions of versioning– Country, Language, Major Rev, Minor Rev, Patchlevel, Feature
set• Sales force lack knowledge of cost• Huge maintenance problem
Modern SE
• Packaged components available on the web
• The dream of reuse being realized
• For large projects, the nightmare of reuse being realized– n-dimensions of versioning
• Still, largely w/in control– Can choose when to include the latest jar
The Web Ontology Analogy
• Packaged components available on the web
• The dream of reuse being realized
• For large projects, the nightmare of reuse being realized– n-dimensions of versioning
• If using imports– CanNOT choose when to include the latest
owl:Ontology & Namespaces
• No Semantics– In a real sense, not part of the language– Imports, versioning
• What is an ontology?
• Not packaging mechanisms– Yet used that way
Imports for “Layering”
Upper Ontology
OWL-Time
Fluents Ontology
Events Objects
App-specific view
Imports for Language Levels
RDFS- Axioms
OWL-DL Axioms
OWL-Full Axioms
Key Observation
• OWL&RDF are axiom-based languages– not frame-based or object-oriented
• The definition of a class or property is not in one place (despite some tools)
(Class cdo:CarsDomainObject)(Class cdo:Car partial cdo:CarsDomainObject)
(Class rdo:RacingDomainObject(Class cdo:Car partial rdo:CarsDomainObject)
Separating axioms by language(Class RigidClass partial (restriction oc:subClassOf allValuesFrom (complementOf(AntiRigidClass)))(Class NonRigidClass partial)(disjointClasses RigidClass NonRigidClass)
(rdfs:subClassOf RigidClass owl:Class)(rdfs:subClassOf NonRigidClass owl:Class)(sameAs oc:subClassOf rdfs:subClassOf)
(RigidClass cdo:Car)
The Dark Side
(Class oc:RigidClass partial restriction oc:subClassOf allValuesFrom complementOf(oc:AntiRigidClass))(Class oc:NonRigidClass partial)(disjointClasses oc:RigidClass oc:NonRigidClass)
(Class oc:RigidClass partial oc:nonRigidClass)
What does it mean?????
Wherefore Reasoning?
• “Glorified Compiler”• Build a taxonomy [Rector]• …
• The “user community” is still unsure what the purpose of reasoning is
A looming problem
• Prediction– Ontology maintenance will become the
significant problem as ontologies become more mainstream
– Will follow the SE model (80% of cost)
• Observation/Conjecture– High quality ontologies are easier to maintain
Software Maintenance
• Fixing Bugs
• Testing
• Enhancing
Ontology Maintenance
• Fixing Bugs– Inconsistent– Inaccurate– Inefficient
• Testing– Regression tests– Test Suites– Meta tag sets for test
content– Ablation tests
• Enhancing– Tweaking
• Richness• Correctness• Organization• Meta-level consistency• Efficiency
– Extending• Improving coverage• Extending commitment• Integration
– Refactoring
Of Chickens and Eggs
• Many other fields focus on large information artifacts– DB, DL, SE
• Other fields of information processing have hit a “wall”– IR, NLP, semantic integration
• Guess where they’re looking for help?