Extração e explicitação de relações semânticas para a representação do conhecimento de documentos acadêmicos: um estudo de caso a partir de uma estrutura classificatória

Abstract

Semantic relations are fundamental for understanding the nature of the connection between two concepts in a domain. For academic documents, such as dissertations and theses, the representation of the knowledge inherent to them, with explicit semantic relations, supports user comprehension of occasionally complex studies in a specific domain. Thus, the aim of this thesis is to propose a model of semantic relations extraction for knowledge representation of academic documents in Brazilian Portuguese. For the formulation of the model, a bibliographical survey about the subjects that compose the theme of the thesis was carried out. One of the outcomes was a compilation of semantic relation classifications of several authors, thus composing a taxonomy of semantic relations in Portuguese. A literature review was carried out as well, and it pointed to the lack of research about extraction of semantic relations in the national scenario. In this context, a web information system called Semantizar was developed, in order to support the extraction of semantic relations from classificatory structures that represent specific academic documents. Thus, Semantizar performs, in these documents, searches for concept pairs in sentences, indicating the existence of a semantic relation. A case study was executed, in order to evaluate the extraction and explanation of semantic relations through Semantizar; in this sense, the system has brought relevant contributions to the research on the extraction of semantic relations. The results reveal that when two concepts of a classificatory structure exist in a sentence, a semantic relation between them can, in fact, exist. Another significant contribution was the discovery of a new subtype of associative semantic relation, which was named agent-subordinate. Finally, the conclusion is that this study is relevant to the scientific community, mainly because it introduces important findings for the extraction of semantic relations for knowledge representation of academic documents, which can be applied in the Brazilian scenario.