Avaliação de ontologia com base na comparação a um corpus

Abstract

The guiding aim of this research was to evaluate a domain ontology through an evaluation proposal based on comparison to a text corpus. The ontology proposed to evaluate in this study was carried out within the EMBRAPA Campinas, it is called OntoAgroHidro and represents the impact of climate changes on agriculture and water resources. The selection of this theme is based on problems related to the existenceof inefficient ontologies, ie, that do not represent properly the domain that propose to represent. Therefore, it is believed that the exposed problems can be minimized with the thematic ontology evaluation, which proposes to verify the suitability of the ontology in the representation of particular domain of knowledge. The research is characterized as exploratory, qualitative and applied, starting with the exploration of the relevanttopics to the research in the Information Science and Computer Science literatures. The theoretical and methodological foundation permeated concepts on topics such as organization and knowledge representation, definitions of term and concept, ontologies and competency questions. The literature review covered the thematic ontology evaluation. The research methodology was based on a proposal for a data-drivenontology evaluation developed by Brewster et al (2004). The proposal was adapted and has four steps: 1) definition of competency questions; 2) selection of the collection of documents that compose the corpus for the ontology evaluation; 3) selection of expressions related to the concepts of the competency questions; 4) evaluation and inclusion of expressions in the ontology. The results demonstrated that the evaluatedontology presented a rate of approximately 70% of the text corpus concepts selected for the evaluation. This shows the need to improve the representation of some concepts in the ontology. The results also indicate particularities in the representation of concepts by OntoAgroHidro, indicating that concepts related to water resourcesdomain had a better representativeness. We concluded that the applied ontology evaluation was satisfactory, as identified several areas for improvement in the evaluated ontology.