Design and Evaluation of Mid-level Ontologies

点击翻译按钮获取中文摘要

Many applications in the domains of e-commerce, e-government, and e-science require the semantic integration of a wide variety of datasets and semantic interoperability among diverse software systems and services. Although domain ontologies have been proposed for decision support and semantic integration, a number of challenges remain:- Within the same domain, there are multiple conflicting ontologies, without a clear understanding of the metalogical relationships between them.- The design of an ontology is hindered because the intended models are difficult to characterize, and even with this characterization, the class of intended models can be difficult to axiomatize. - Weak axiomatizations of an ontology allow the existence of unintended models; this is problematic for decision support systems insofar as there will exist sentences that areentailed by all intended models of the ontology but which are not provable from the axioms.- The lack of modularization inhibits the reusability of large ontologies.The content of COLORE now allows us to address fundamental problems in ontology design and evaluation. COLORE explicitly specifies the metatheoretic relationships among ontologies, such as conservative extension, nonconservative extension, and definable interpretation. This allows one to explicitly determine common subtheories that areshared by multiple ontologies in the same domain. Furthermore, these relationships can be used to verify that the intended models of an ontology are equivalent to the models of the axiomatization, and this verification can be used to modularize the ontology.Upper ontologies such as Cyc and SUMO respond the above challenges by defining concepts which are specializations of generic concepts and generalizations of domain ontologies. Since repositories contain many generic ontologies, one possible approach is to design ontologies which are intermediate between the domain and generic ontologies.Nevertheless, there has been no formal characterization of such midlevel ontologies.To address these challenges, this project has the following objectives:1. Specify the metalogical relationships (used in COLORE) among existing domain ontologies.2. Design new modular midlevel ontologies by merging generic ontologies and then formally evaluate them with respect to their logical and mathematical properties.3. Design midlevel ontologies by definitional extensions of generic ontologies.4. Specify techniques for the modularization of domain ontologies based on their metalogical relationships.5. Implement a software environment to support ontology design and evaluationThe following domain ontologies will be in the scope of this project: GoodRelations, FIBO, Units of Measure, Date-Time Vocabulary, SCRIBE, Foundational Model of Anatomy, BioPAX, together with the various ontologies for events, addresses, and products, which are used in Semantic Web applications.