Similarity-Based Ontology Alignment in OWL-Lite

Interoperability of heterogeneous systems on the Web will be admittedly achieved through an agreement between the underlying ontologies. However, the richer the ontology description language, the more complex the agreement process, and hence the more sophisticated the required tools. Among current ontology alignment paradigms, similarity-based approaches are both powerful and flexible enough for aligning ontologies expressed in languages like OWL. We define a universal measure for comparing the entities of two ontologies that is based on a simple and homogeneous comparison principle: Similarity depends on the type of entity and involves all the features that make its definition (such as superclasses, properties, instances, etc.). One-to-many relationships and circularity in entity descriptions constitute the key difficulties in this context: These are dealt with through local matching of entity sets and iterative computation of recursively dependent similarities, respectively. 1 The ontology alignment problem Ontologies are seen as the solution to data heterogeneity on the web. However, the available ontologies could themselves introduce heterogeneity: given two ontologies, the same entity can be given different names or simply be defined in different ways, whereas both ontologies may express the same knowledge but in different languages. Semantic interoperability can be grounded in ontology reconciliation. The underlying problem, which we call the “ontology alignment” problem, can be described as follows: given two ontologies each describing a set of discrete entities (which can be classes, properties, predicates, etc.), find the relationships (e.g., equivalence or subsumption) that hold between these entities. Alignment results can further support visualization of correspondences, transformation of one source into another or formulation of bridge axioms between the ontologies. An overview of alignment methods is presented in §2. We focus on automatic and autonomous ontology alignment, although more interactive scenarios may be built on top of the proposed technique (e.g., complete a partial alignment or use the result as a suggestion to the user). It will be also assumed that the ontologies are described within the same knowledge representation language: OWL-Lite (§3.1). The language is based on various features: classes and subsumption, properties and type constraints, etc. Instead of using the external textual form, we define a dedicated typed graph representation of the language that concentrates the necessary information for computing the similarity between OWL entities (§3.2). This paper defines a similarity measure that encompasses all OWL-Lite features (§4.1) while overcoming key alignment problems such as collection comparison (§4.2) and circularities (§4.3). 1 This work has been partially supported by grants from the French consulate in Montréal and the Centre Jacques Cartier. First author is funded in part by project IST-2004-507482 (Knowledge web). 2 INRIA Rhône-Alpes,Jerome.Euzenat@inrialpes.fr 3 Université de Montréal, Petko.Valtchev@umontreal.ca Our approach is based on previous work on object-based knowledge representation similarity [15] which is here adapted to the alignment within current web languages. The computed measure is then used for generating an actual alignment (§5).