130 likes | 217 Views
QA of Ontologies. OWL Tutorial December 6 th 2005. The GIGO Problem. Logic may be necessary But it is not sufficient Can not validate all possible inferences in advance Instead must prove: Reasoner is sound Axioms are correct Induction does the rest. Barriers to Ontology QA.
E N D
QA of Ontologies OWL Tutorial December 6th 2005
The GIGO Problem Logic may be necessary But it is not sufficient Can not validate all possibleinferencesin advance Instead must prove:Reasoner is sound Axioms are correct Induction does the rest
Barriers to Ontology QA • Absolute Measures • No ‘Gold Standard’ • Mutual cross validation only as good as the parts • Manual checking error-prone • and can’t measure HOW error prone because of (1) • Comparitive Measures • Better than worse does not imply good • Relative Measures • Provides unequivocal evidence of improvement • But not of proximity to goal • Falling error detection rate does not imply none exist
Types of Axiom Quality Philosophical Rigour Ontological commitment Content correctness Fitness for purpose
Philosophical Rigour • 2500 years of research • Theories of time, mereonomy, containment • Often FOL, so not computable • Similar upper level ontologies • DOLCE, BFO • But not 100% agreement: Realist vs Cognitivist
Ontological commitment • Formally specified semantic equivalence • Logical transformation to canonical form • Semantically equivalent but no logical transform • ‘Fixation of femur by means of inserting pins’ • ‘Insertion of pins to fixate the femur’ • Metamodel rules/commitments • Arbitrary choice of preferred form • Conventions to be applied throughout ontology • And by all applications that use it
Content correctness • Metadata Provenance, lexical annotations etc • Truth • ‘Structure of labial vein’ is-a ‘Superficial vein of face’ • Completeness – ambiguity and omission • Thymus secretes Thymosin; Thymosin is-a Hormone…but omits Thymus is-a endocrine gland • Conciseness • Redundant inclusion of inferrable axioms • Consistency – contradictory, duplicated, circular • endocrine surgery vs endocrine surgeons • Traumatic unilateral amputationUnilateral traumatic amputation
Fitness for Purpose • Best theories no guarantee of usability or utility • Lab experiences no predictor of field behaviour • All for nothing, if user can’t use it • Interrater variability
X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X Fitness for Purpose:Inter-rater variability Headcloth Cloth Scarf Model Person Woman Adults Standing Background Brown Blue Chemise Dress Tunics Clothes Suitcase Luggage Attache case Brass Instrument French Horn Horn Tuba
Fitness for Purpose:Inter-rater variability • Miscoding • Code meaning is inappropriate to thing being described • Instrument definitely not a french horn • Missed coding • Not coding something that could be coded • No code for the table/platform • Overcoding • Code meaning is more detailed than justified • Can the gender really be determined? • Undercoding • Code meaning is less detailed than justified • Brass Instument vs Tuba
Fitness for PurposeWhat ontological properties.. • Increase usablility and utility? • Are a prerequisite for them? • Decrease usability or utility?
Ontology QAHow much quality do we need? Perfection is unattainable Trade-offs between quality and… Performance Cost Maintainability Usability Acceptance Utility
REVIEWExploreconsequences DEVISEQIAlgorithm APPLYQIAlgorithm Need for CQI, not final QAAssure the process not the product ASSESSIdentifyproblems Use/Test cases & exemplars