Comparing Two Approaches to Eliminating Cycles in Terminological Systems.

Mougin F, Bodenreider O

Proc RFIA 2006


Terminological resources often have some hierarchical organization. Therefore, systems integrating multiple terminologies have a polyhierarchical organization and may exhibit conflicting hierarchical relations, leading to cycles in the graph. Our study presents some of the causes of cycles in the UMLS Metathesaurus graph, a system integrating more than 100 biomedical vocabularies. We compare two approaches to eliminating these cycles. The naive approach simply avoids loops during the graph traversal by preventing nodes from being visited more than once. The formal approach eliminates cycles from the graph a priori, using a set of rules and heuristics. An application of terminological graphs is used in this comparison: computing the set of descendants for a given concept. The naive approach is easier to implement, but does not ensure thatinappropriate semantic relations are eliminated. In contrast, the formal approach improves the semantic consistency in sets of descendants but is more complex and requires domain expertise. Finally, although semantic consistency is better with the formal approach, it remains suboptimal due to limitations of the UMLS itself.

Mougin F, Bodenreider O. Comparing Two Approaches to Eliminating Cycles in Terminological Systems. 
Proc RFIA 2006