A Generalization of the Winkler Extension and its Application for Ontology Mapping

Mapping ontologies is a crucial process when facilitating system interoperability and information exchange. Ontology Mapping systems commonly utilize string metrics in the mapping process to compare concept names. String metrics can be extended using the Winkler method, which increases the similarity value of two strings if these have a common prefix. A common occurrence for two corresponding ontology concepts is that the name of the first concept is a non-prefix sub-string of the name of the second concept. The Winkler extension does not allocate a higher similarity value to these pairs of strings, however intuitively this indicates that the two names have a similar meaning. This paper proposes a generalization of the Winkler extension, such that pairs of names with large common non-prefix sub-strings receive a higher similarity value as well. The proposed metric is evaluated on a record-matching dataset and a dataset from the Ontology Alignment Evaluation Initiative. The experiments reveal that metrics applying our proposed generalization outperform the same metrics when applying the Winkler extension.