In this paper we present the evaluationof a set of string similarity metrics usedto resolve the mapping from strings toconcepts in the UMLS MetaThesaurus.String similarity is conceived as a singlecomponent in a full Reference ResolutionSystem that would resolve such a mapping.Given this qualification, we obtainpositive results achieving 73.6 F-measure(76.1 precision and 71.4 recall) for thetask of assigning the correct UMLS conceptto a given string. Our results demonstratethat adaptive string similarity methodsbased on Conditional Random Fieldsoutperform standard metrics in this domain.
展开▼