Ascendant Hierarchical Clustering for Instance Matching

Abstract

With the rapid advancement of semantic web, especially of the web of data, a growing number of independently designed and structured datasets represented by ontologies, are built and need to be integrated in the Linked Open Data (LOD) cloud. In this context, instance matching is presented as a fundamental solution for ontological data sharing and integration. It aims to link co-referent instances (instances that refer to the same real world objects) from various datasets to allow them to complement each other. Traditional systems depend a lot on the quality of schema-level mappings and especially on property mappings, which are not always obvious for the LOD paradigm. In this paper, we propose a schema-free instance matching approach that is independent from property matching results. We transform the instance matching problem into a clustering problem and we solve it by Ascendant Hierarchical Clustering algorithm. Furthermore, we employ some structural information to filter-out obtained results and eliminate wrong mappings. We evaluate our approach on instance matching track from Ontology Alignment Evaluation Initiative (OAEI) benchmark. The experiments show that our approach gets prominent results compared to several participating systems in OAEI’2019 and OAEI’2020.