Lattice-based confrontation of ontologies

点击翻译按钮获取中文摘要

Interoperability of heterogeneous systems on the Web, e.g., two applications exchanging electronic documents, will be admittedly achieved through a confrontation and the subsequent agreement between the underlying ontologies. The most basic confrontation problem, called ontology alignment, reads as follows: given two ontologies each describing a set of entities (concepts, relations), find the relationships (e.g., equivalence or generality) that hold between these entities. Most alignment methods use either similarity or machine learning as tool for detecting plausible alignments and only a small number rely on semantic-based reasoning. They all suffer on specific shortages, ranging from inefficiency to output limitations. As a remedy we shall investigate a novel alignment approach, based on the joint use of concept lattices and similarity, the former as support structure for alignment hypothesis generation and the latter as a ranking mechanism for hypotheses. Concept lattices provide a hierarchical and contextual representation of ontology entity similarities based on shared binary attributes. By binding those attributes to the structural and the terminological aspects of entity descriptions, we expect to approximate the existing semantic relationships between entities and make them appear in the concept lattice as distinctive "patterns". Similarity will then be used to filter the most plausible instances of such patterns before turning them into alignment cells. The ultimate goal is the design of an operational framework for ontology confrontation based on lattices and similarity.