Relevance-based evaluation of alignment approaches: The OAEI2007 food task revisited
TNO Industrie en Techniek
van Hage, W.R.
Current state-of-the-art ontology-alignment evaluation methods are based on the assumption that alignment relations come in two flavors: correct and incorrect. Some alignment systems find more correct mappings than others and hence, by this assumption, they perform better. In practical applications however, it does not only matter how many correct mappings you find, but also which correct mappings you find. This means that, apart from correctness, relevance should also be included in the evaluation procedure. In this paper we expand the sample-based evaluation of the OAEI 2007 food task with a sample evaluation that uses relevance to prototypical search tasks as a selection criterion for the drawing of sample mappings.
To reference this document use:
3rd International Workshop on Ontology Matching, OM-2008 - Collocated with the 7th International Semantic Web Conference, ISWC-2008, 26 October 2008 through 26 October 2008, Karlsruhe, 431 (431), 234-238
CEUR Workshop Proceedings