Catálogo de publicaciones - libros

An ontology typically provides a vocabulary describing a domain of interest and a specification of the meaning of terms in that vocabulary. Depending on the precision of this specification, the notion of ontology encompasses several data or conceptual models, e.g., classifications, database schemas, fully axiomatised theories. Ontologies tend to be everywhere. They are viewed as the silver bullet for many applications, such as database integration, peer-to-peer systems, e-commerce, semantic web services, social networks [Fensel, 2004]. They are, indeed, a practical means to conceptualise what is expressed in a computer format [Brodie , 1984]. However, in open or evolving systems, such as the semantic web, different parties would, in general, adopt different ontologies. Thus, merely using ontologies, like using XML, does not reduce heterogeneity: it raises heterogeneity problems to a higher level.

Matching models is an important operation in traditional applications, such as ontology integration, schema integration, or data warehouses. Typically, these applications are characterised by heterogeneous structural models that are analysed and matched either manually or semi-automatically at design time. In such applications matching is a prerequisite of running the actual system.

In a distributed and open system, such as the semantic web and many other applications presented in the previous chapter, heterogeneity cannot be avoided. Different actors have different interests and habits, use different tools and knowledge, and most often, at different levels of detail. These various reasons for heterogeneity lead to diverse forms of heterogeneity, and, therefore, should be carefully taken into consideration.

Having defined what the matching problem is, we attempt at classifying the techniques that can be used for solving this problem. The major contributions of the previous decades are presented in [Larson , 1989, Batini , 1986, Kashyap and Sheth, 1996, Parent and Spaccapietra, 1998], while the topic through the recent years have been surveyed in [Rahm and Bernstein, 2001, Wache , 2001, Kalfoglou and Schorlemmer, 2003b]. These three works address the matching problem from different perspectives (artificial intelligence, information systems, databases) and analyse disjoint sets of systems. [Shvaiko and Euzenat, 2005] have attempted to consider the above mentioned works together, focusing on schema-based matching methods, and aiming to provide a common conceptual basis for their analysis. Here, we follow and extend this work on classifying matching approaches and use it in the following chapters for organising the presentation.

The goal of ontology matching is to find the relations between entities expressed in different ontologies. Very often, these relations are equivalence relations that are discovered through the measure of the similarity between the entities of ontologies.

The basic techniques presented in Chap. 4 are the building blocks on which a matching solution is built. Once the similarity or (dis)similarity between ontology entities are available, the alignment remains to be computed. This involves more global treatments. In particular, the following aspects of building a working matching system are considered in this chapter:

This chapter is an overview of matching systems which have emerged during the last decade. There have already been some comparisons of matching systems, in particular in [Parent and Spaccapietra, 2000, Rahm and Bernstein, 2001, Do , 2002, Kalfoglou and Schorlemmer, 2003b, Noy, 2004a, Doan and Halevy, 2005, Shvaiko and Euzenat, 2005]. Our purpose here is not to compare them in full detail, though we give some comparisons, but rather to show their variety, in order to demonstrate in how many different ways the methods presented in previous chapters have been practically exploited.

The increasing number of methods available for ontology matching suggests the need for evaluating these methods. However, very few extensive experimental comparisons of algorithms are available. Matching systems are difficult to compare, but we believe that the ontology matching field can only evolve if evaluation criteria are provided. These should help system designers to assess the strengths and weaknesses of their systems as well as help application developers to choose the most appropriate algorithm.

Once matching is performed, the resulting alignments are usually used in a wider context than a matching system itself. To that extent, several proposals have been made for representing the alignments and exchanging them among tools. This chapter is concerned with these topics.

Matching systems may produce effective alignments that may not be intuitively obvious to human users. In order for users to trust the alignments, and thus use them, they need information about them, e.g., they need access to the sources that were used to determine semantic correspondences between ontology entities. Explanations are also useful when matching large applications with thousands of entities, e.g., business product classifications, such as UNSPSC and eCl@ss. In such cases, automatic matching solutions will find many plausible correspondences, and hence user input is required for performing cleaning-up of the alignment. Finally, explanations can also be viewed and applied as argumentation schemas for negotiating alignments between agents.