Processamento de conhecimento impreciso combinando raciocínio de ontologias fuzzy e sistemas de inferência fuzzy

Abstract

In Computer Science, ontologies are used for knowledge representation in a number of applications, aiming to structure and handle domain semantics through models shared by humans and computational systems. Although traditional ontologies model semantic information and support reasoning tasks, they are based on a formalism which is less suitable to express the vagueness inherent in real-world phenomena and human language. To address this issue, many proposals investigate how traditional ontologies can be extended by incorporating concepts from fuzzy sets and fuzzy logic, resulting in fuzzy ontologies. In special, combining the formalism from fuzzy ontologies with fuzzy rule-based reasoning, which has been successfully applied in the context of fuzzy inference systems, can lead to more expressive inferences involving imprecision. In this sense, this doctoral thesis aims at exploring the integration of fuzzy ontology reasoning with fuzzy inference systems, resulting in the definition and the development of two approaches: HyFOM (Hybrid integration of Fuzzy Ontology and Mamdani reasoning) and FT-FIS (Fuzzy Tableau and Fuzzy Inference System). HyFOM is based on a hybrid architecture combining reasoners for ontologies, fuzzy ontologies and fuzzy inference systems, focusing on the interaction among its independent components. FT-FIS defines an interface between a fuzzy tableau-based algorithm and a fuzzy inference system, including the fuzzyRuleReasoning predicate that allows fuzzy rule-based reasoning to be invoked whenever necessary for fuzzy ontology reasoning tasks. The main contribution of HyFOM and FT-FIS comes from their reasoning architectures, which combine flexibility in terms of fuzzy rule semantics with the collaboration between inferences from both types of reasoning. Experiments regarding the recommendation of touristic attractions, based on synthetic data, revealed that HyFOM and FT-FIS provide integrated inferences, in addition to a more expressive approximation of the relation defined by fuzzy rules than the results from the fuzzyDL reasoner. In experiments involving the evaluation of chemical risk in food samples, based on real data, results obtained by HyFOM and FT-FIS are also more precise than fuzzyDL results, in comparison with reference values available in this domain.