Matching Linked Open Data Entities to Local Thesaurus Concepts
Peter Wetz, Hermann Stern, Jürgen Jakobitsch, Viktoria Pammer‐Schindler · 2012
Abstract. We describe a solution for matching Linked Open Data (LOD) entities to concepts within a local thesaurus. The solution is currently integrated into a demonstrator of the PoolParty thesaurus management software. The underlying motivation is to support thesaurus users in linking locally relevant concepts in a thesaurus to descriptions available openly on the Web. Our concept matching algorithm ranks a list of potentially matching LOD entities with respect to a local thesaurus concept, based on their similarity. This similarity is calculated through string matching algorithms based not only on concept and entity labels, but also on the “context ” of concepts, i.e. the values of properties of the local concept and the LOD concept. We evaluate over 41 different similarity algorithms on two test-ontologies with 17 and 50 concepts, respectively. The results of the first evaluation are validated on the second test-dataset of 50 concepts in order to ensure the generalisability of our chosen similarity matches. Finally, the overlap-, TFIDF- and SoftTFIDFsimilarity algorithms emerge as winners of this selection and evaluation procedure.