Metadata-Driven Semantic Coordination

Abstract

Reuse and combination of disparate datasets on the Semantic Web require semantic coordination, i.e. the ability to match heterogeneous semantic models. Systematic evaluations raised the performance of matching systems in terms of compliance and resource consumption. However, it is equally important to be able to identify diverse matching scenarios, covering a range of variations in the datasets such as different modeling languages, heterogeneous lexicalizations, structural differences and to be able to properly handle these scenarios through dedicated techniques and the exploitation of external resources. Furthermore, this should be achieved without requiring manual tinkering of low-level configuration knobs. As of the Semantic Web vision, machines should be able to coordinate and talk to each other to solve problems. To that end, we propose a system that automates most decisions by leveraging explicit metadata regarding the datasets to be matched and potentially useful support datasets. This system uses established metadata vocabularies such as VoID, Dublin Core and the LIME module of OntoLex-Lemon. Consequently, the system can work on real-world cases, leveraging metadata already published alongside self-describing datasets.