TY - JOUR
T1 - A feature selection approach for anchor evaluation in ontology mapping
AU - Schadd, Frederik C.
AU - Roos, Nico
PY - 2014
Y1 - 2014
N2 - Computing alignments between ontologies is a crucial task for the facilitation of information exchange between knowledge systems. An alignment is a mapping consisting of a set of correspondences, where each correspondence denotes two ontology concepts denoting the same information. In this domain, it can occur that a partial alignment is generated by a domain expert, which can then be exploited by specialized techniques. In order for these techniques to function as intended, it must be ensured that the given correspondences, also known as anchors, are indeed correct. We propose an approach to this problem by reformulating it as a feature selection task, where each feature represents an anchor. The feature space is populated with a set of reliably generated correspondences, which are compared with the anchors using a measure of alignment. We apply feature selection techniques to quantify how well the anchors align with this set of correspondences. The resulting scores are used as anchor reliability measures and combined with the anchor similarities.\nWe evaluate the approach by generating a set of partial alignments for the used dataset and weighting the concept similarities with anchor evaluation measure of our approach. Three different similarity metrics are used, a syntactic, structural and semantic metric, in order to demonstrate the effectiveness of our approach.
AB - Computing alignments between ontologies is a crucial task for the facilitation of information exchange between knowledge systems. An alignment is a mapping consisting of a set of correspondences, where each correspondence denotes two ontology concepts denoting the same information. In this domain, it can occur that a partial alignment is generated by a domain expert, which can then be exploited by specialized techniques. In order for these techniques to function as intended, it must be ensured that the given correspondences, also known as anchors, are indeed correct. We propose an approach to this problem by reformulating it as a feature selection task, where each feature represents an anchor. The feature space is populated with a set of reliably generated correspondences, which are compared with the anchors using a measure of alignment. We apply feature selection techniques to quantify how well the anchors align with this set of correspondences. The resulting scores are used as anchor reliability measures and combined with the anchor similarities.\nWe evaluate the approach by generating a set of partial alignments for the used dataset and weighting the concept similarities with anchor evaluation measure of our approach. Three different similarity metrics are used, a syntactic, structural and semantic metric, in order to demonstrate the effectiveness of our approach.
U2 - 10.1007/978-3-319-11716-4_14
DO - 10.1007/978-3-319-11716-4_14
M3 - Article
SN - 1865-0929
VL - 468
SP - 160
EP - 174
JO - Communications in Computer and Information Science
JF - Communications in Computer and Information Science
ER -