Alignment Evoluttion under Ontology Change : A formal Framework and Tools

Abstract

Alignment of ontologies is the backbone of semantic interoperability. It facilitates the import of data from an ontology to another, translating queries between them, or merging ontologies in a global one. However, these services cannot be guaranteed throughout the life cycle of the ontology. The problem is that the evolution of aligned ontologies may affect and make obsolete the alignment. Contributions of this dissertation touch the literature review and the methodology knowledge sides of the alignment evolution problem. At the methodology knowledge side, the dissertation proposes a formal framework that consists of a number of phases, each having a specific purpose. The framework facilitates ontology change identification for maintainers. On the light of base revision theory, the framework presents a set of generic operators for evolving alignments from a consistent state to another consistent state with a minimal of change. The framework adapts the Hitting set algorithm of diagnosis theory to concretize these operators. Besides, the framework is extended with a global method which is an orchestration of a set of operations each of which is designed to take care of one aspect of the alignment change process. Finally, the framework allows to maintainers reviewing the change before implementation. In what concerns the literature review side, the dissertation mentions the importance of the problem and recommends the separation of its study from the study of the ontology evolution problem. Besides, the dissertation suggests classifying the alignment evolution approaches in two categories. Approaches of the former are corrective since they check and resolve inconsistences after change while approaches of the latter are adaptive and perfective since they only adapt the alignment according to detected changes in ontologies. Moreover, the dissertation demonstrates the advantage of the proposed approach relatively to others. The results show that neither ontology matching nor alignment debugging methods fit well for the alignment evolution problem.