A Neural-Networks-Based Approach for Ontology Alignment

Abstract

Ontologies are key elements in the Semantic Web for providing formal definitions of concepts and relationships. Such definitions are needed to have data that could be understood and reasoned upon by machines as well as humans. However, because of the possibility of having many Ontologies in the web, alignment – which aims providing mappings across them – is a necessary operation. Many metrics have been defined for ontology alignment. The so-called simple metrics use linguistic or structural features of Ontological concepts to create mappings. Compound metrics, on the other hand, combine some of the simple metrics to have a better results. This paper reports our new method for compound metric creation. It is based on a supervised learning approach in data mining where a training set is used to create a neural network model, performs sensitivity analysis on it to select appropriate metrics among a set of existing ones, and finally constructs a neural network model to combine the result metrics into a compound one. Empirical results of applying it on a set of Ontologies is also shown in this paper.