Animating the reference terminology – showing classifiers at work
Ed Cheetham, Principal Terminology Specialist
Introduction
• In addition to being hand-curated, SNOMED CT’s content is also (re-)organised, and its development quality assured, by the use of a description logic (DL) classifier.
What is description logic? [Spackman 2008]
•Mathematical viewpoint:A family of logics characterized by Formal set-theoretic semantics
Proofs of correctness and completeness of computationProofs of algorithmic complexity (PSpace, NP-complete,
NExpTime, etc)
•Knowledge representation viewpoint:A set of constructs for representing terminological knowledgeAlgorithms and their implementations for performing:
Subsumption (testing pairs of expressions to see whether one is a subtype of the other & vice versa)
Classification (structuring a set of expressions according to their subsumptionrelationships)
DL-based classification – simply put...
• Agree ‘set of constructs’ [operators, roles]• Make certain properties of content formally explicit
‘Stated relationships
• Decide whether content is sufficiently defined in such terms
Fully defined/primitive• ‘Run’ classifier
Defined – what are kinds of ‘me’? What am I a kind of?Primitive – what am I a kind of?
Appendectomy Is_A Excision procedure ANDHas_method=Excision ANDHas_site=Appendix
AND, OR, NOT, Roles, Role hierarchies
Appendectomy Fully definedOperation GI tract Fully definedExcision, Appendix Primitive
Total Is A RoleStated 778435 525350 253085Inferred 1035196 611737 423459
Protégé: http://protege.stanford.edu/Use does not indicate endorsement, but extremely valuable to illustrate points discussed
Gephi: http://gephi.org/Use does not indicate endorsement, but extremely valuable to illustrate points discussed
Blue lines = stated and inferred
Red lines = stated (removed as redundant)
Green lines = inferred
Conclusions
•DL-based classification is an intrinsic part of SNOMED CT development
Necessary QA feature of large KR product•Based on ‘what it is told’ and the expressivity of the other ‘constructs’, content is ruthlessly reorganised
New ‘inferred’ knowledge (reclassification)Sometimes intended, sometimes unintended