Strategies and Methods for Ontology Alignment

Introduction

The concept of ontologies has been widely applied as a means for providing a shared understanding of common domains. However, the generalized use of large distributed environments, such as WWW, has resulted in the proliferation of different ontologies, even in the same or similar domain, calling for methods to align the diverse ontologies. As the number of ontologies available increases, the need to align these ontologies increases (Y. Kalfoglou & M Schorlemmer, 2003). A single ontology is not sufficient to support the distributed nature of the semantic web or the operations of many businesses (Thomas, Redmond, & Yoon, 2009) or governments (Santos & Madeira, 2010). In order to effectively leverage knowledge represented in these overlapping ontologies, ontology alignment may be useful. Ontology alignment is achieved in part by mapping the concepts of one ontology to the concepts of another ontology. Ontology alignment may also be a step in the larger process of ontology integration or merging (Pinto, Gomez-Perez, & Martins, 1999).

Ontologies may be manually mapped by a knowledge engineer or domain expert; however, this process has shortcomings. Manual mapping is tedious, error prone, and hinders ontology maintenance (Ding & Foo, 2002). Automated mapping may occur at the schema-level or the instance-level. Schema-level mapping begins with two schemas as input producing output in the form of semantic links between the elements of the input schemas (Rahm & Bernstein, 2001). These links may range from simple one-to-one links to complex many-to-many, semantically labeled links (Embley, Xu, & Ding, 2004). Instance-level mapping occurs by using classified instance data in order to construct links between concepts based on the co-occurrences of the instances (Isaac, Van der Meij, Schlobach, & Wang, 2007). Finally, a hybrid approach employs schema and instance techniques.

This chapter presents a hybrid, ontology alignment system, called ALIGN. Two crucial steps in this system are:

• •

  A schema-based method to determine the similarity between a concept in one ontology and a concept in another ontology based on the lexical match of the two. This is considered a schema-level method in that it employs the concept category of each ontology, although it does not make inferences based on larger structures in either ontology.

• •

  An instance-based approach to determine the similarity between two concepts based on their instances using a Jaccard coefficient similarity measure.

Notably, this work illustrates how several, readily available Semantic Web technologies can be used for ontology alignment.

The system that is developed could be applied to any two ontologies that maintain the class-instance relationship. To demonstrate the system, the authors have constructed two adverse drug reaction ontologies. A set of tests were then conducted on the extent to which the system could successfully exploit the information in both ontologies through mapping. The following sections present related work, the system architecture, a detailed development methodology and an application, and the conclusions.

Background

As previously noted, the interest in ontologies long predates the current interest in the World Wide Web and the power of ontologies to help people organize and retrieve information. Instead, one might go back, at least, three thousand years to the origins of metaphysics (see, for instance, Thales of Miletus, the pre-Socratic Greek philosopher) and see that philosophers were pondering the relations between particulars and universals, and those relations are the essence of ontologies. Jumping to the computerized world, one can find innumerable historical precedents to the interest in ontology alignment. One such path through the history is sketched next.