Reasoning by Analogy in the Generation of Domain Acceptable Ontology Refinements

Refinements generated for a knowledge base often involve the learning of new knowledge to be added to or replace existing parts of a knowledge base. However, the justifiability of the refinement in the context of the domain (domain acceptability) is often overlooked. The work reported in this paper describes an approach to the generation of domain acceptable refinements for incomplete and incorrect ontology individuals through reasoning by analogy using existing domain knowledge. To illustrate this approach, individuals for refinement are identified during the application of a knowledge-based system, EIRA; when EIRA fails in its task, areas of its domain ontology are identified as requiring refinement. Refinements are subsequently generated by identifying and reasoning with similar individuals from the domain ontology. To evaluate this approach EIRA has been applied to the Intensive Care Unit (ICU) domain. An evaluation (by a domain expert) of the refinements generated by EIRA has indicated that this approach successfully produces domain acceptable refinements.