Analysing Syntactic Regularities and Irregularities in SNOMED-CT.

Area of Science:

• Computer Science
• Artificial Intelligence
• Ontology Engineering

Background:

• Ontology quality assurance is crucial for real-world applications but remains challenging, especially for undocumented ontologies.
• Existing methods lack tools for automated detection of syntactic regularities and irregularities in ontology axioms.
• This gap hinders the ability to ensure ontologies comply with established coding standards and patterns.

Purpose of the Study:

• To demonstrate the RIO (Regularity Identification in Ontologies) framework for detecting syntactic regularities in ontologies.
• To apply cluster analysis on ontology entities to identify repetitive axiom patterns.
• To provide a tool for automated ontology quality assurance by detecting matches and deviations from established patterns.

Main Methods:

• Utilized the RIO framework for analyzing OWL ontologies.
• Employed standard machine learning clustering approaches to group similar entities based on their axioms.
• Generalized axiom patterns to identify regularities and deviations within the ontology structure.

Main Results:

• The RIO framework successfully detected regularities and repetitive structures in OWL ontology axioms.
• Analysis of SNOMED-CT modules revealed a high number of regularity deviations, with some patterns followed by only 5% of axioms.
• A subset of deviations were identified as 'design defects,' primarily incomplete descriptions, validating the framework's effectiveness.

Conclusions:

• Automatic detection of regularities and inspection of irregularities in ontologies is feasible.
• The RIO framework serves as a valuable tool for identifying and reporting axiom pattern matches and mismatches for expert evaluation.
• Standard machine learning clustering techniques can significantly contribute to automated ontology quality assurance.