Search

The formal semantics of the Web Ontology Language (OWL) enables automated reasoning overOWL knowledge bases, which in turn can be used for a variety of purposes including knowledgebase development, querying and management. Automated reasoning is usually done by means ofdeductive (proof-theoretic) algorithms which are either provably sound and complete or employ approximate methods to trade some correctness for improved efficiency. As has been argued elsewhere, however, reasoning methods for the Semantic Web do not necessarily have to be based on deductive methods, and approximate reasoning using statistical or machine-learning approaches may bring improved speed while maintaining high precision and recall, and which furthermore may be more robust towards errors in the knowledge base and logical inconsistencies.In this thesis, we show that it is possible to learn a linear-time classifier that closely approximatesdeductive OWL reasoning in some settings. In particular, we specify a method for extracting featurevectors from OWL ontologies that enables the ID3 and AdaBoost classifiers to approximate OWLquery answering for single answer variable queries. Amongst other ontologies, we evaluate ourapproach using the LUBM benchmark and the DCC ontology (a large real-world dataset abouttraffic in Dublin) and show considerable improvement over previous efforts.

This work is structured as follows. In Chapter 1 we brief y introduce some background and we recall the main definitions and results that will be needed. In Chapter 2 we collect several results that show connections between fuzzy relations and fuzzy operators. We also extend some of the known definitions to a more general ones. Chapter 3 is devoted to the analysis of permutability. We start analyzing certain cases of fuzzy relations and fuzzy operators. In particular, fuzzy preorders and indistinguishabilities and fuzzy consequence and interior operators. Then, we relate permutability of fuzzy relations with permutability of the operators that they induce using the different processes defined in Chapter 2. We focus in the cases of fuzzy operators induced by fuzzy preorders and similarities. Finally, we present the conclusions and the possible future work.

This thesis examines the issue of uncertainty reasoning and representation in expert systems. Uncertainty and expert systems are defined. The value of uncertainty in expert systems as an approximation of human reasoning is stressed. Five alternative methods of dealing with uncertainty are explored. These include Bayesian probabilities, Mycin confirmation theory, fuzzy set theory, Dempster-Shafer's theory of evidence and a theory of endorsements. A toolkit to apply uncertainty processing in PROLOG expert systems is developed using fuzzy set theory as the basis for uncertainty reasoning and representation. The concepts of fuzzy logic and approximate reasoning are utilized in the implementation. The toolkit is written in C-PROLOG for the PYRAMID UNIX system at the Rochester Institute of Technology.