UM PROCESSO INDEPENDENTE DE DOMÍNIO PARA O POVOAMENTO AUTOMÁTICO DE ONTOLOGIAS A PARTIR DE FONTES TEXTUAIS

Abstract

Knowledge systems are a suitable computational approach to solve complex problems and to provide decision support. Ontologies are an approach for knowledge representation about an application domain, allowing the semantic processing of information and, through more precise interpretation of information, turning systems more effective and usable. Ontology Population looks for instantiating the constituent elements of an ontology, like properties and non-taxonomic relationships. Manual population by domain experts and knowledge engineers is an expensive and time consuming task. Fast ontology population is critical for the success of knowledge-based applications. Thus, automatic or semi-automatic approaches are needed. This work proposes a generic process for Automatic Ontology Population by specifying its phases and the techniques used to perform the activities on each phase. It also proposes a domain-independent process for automatic population of ontologies (DIAOPPro) from text that applies natural language processing and information extraction techniques to acquire and classify ontology instances. This is a new approach for automatic ontology population that uses an ontology to automatically generate rules to extract instances from text and classify them in ontology classes. These rules can be generated from ontologies of any domain, making the proposed process domain independent. To evaluate DIAOP-Pro four case studies were conducted to demonstrate its effectiveness and feasibility. In the first one we evaluated the effectiveness of phase "Identification of Candidate instances" comparing the results obtained by applying statistical techniques with those of purely linguistic techniques. In the second experiment we evaluated the feasibility of the phase "Construction of a Classifier", through the automatic generation of a classifier. The last two experiments evaluated the effectiveness of DIAOP-Pro into two distinct domains: the legal and the tourism domains. The results indicate that our approach can extract and classify instances with high effectiveness with the additional advantage of domain independence.