Indexação de acórdãos por meio de uma ontologia jurisprudencial populada a partir de um corpus jurídico real

Abstract

Aiming at improving the quality and accuracy of jurisprudential search services of the courts of justice, this dissertation proposes the automatic identification of relevant terms, or sentences, in a corpus of jurisprudential documents. Each identified term should be related to a concept that represents its meaning in the domain of jurisprudence, thus instancing an individual of that domain. At the end, each individual must be stored in knowledge base, thus enabling semantic queries over the documents. To achieve these objectives, methods were proposed for the extraction of relevant sentences and to build an application ontology for the representation of judgments. Furthermore, ontology navigation and search methods were proposed. For the extraction of sentences, several Text Mining techniques were used, such as Sentence Extraction, Analysis of Regular Expressions, Stemming, Stop-words and Controlled Vocabularies. The construction of the ontology followed the methodology OTKM and used some knowledge representation languages, such as DL, RDF and OWL. The Jena framework was applied for navigation in the ontology. SPARQL queries language was applied for search in the ontology. To validate the methods here proposed, an application ontology for the domain of judgments were built, as well as a knowledge management application to the judgments based on the ontology. The application includes routines for knowledge extraction from a corpus of judgments, population of the ontology with the extracted knowledge and then semantic search on the populated ontology. The ontology. The ontology and the knowledge extracted from 50 judgments were submitted to criticism by experts in jurisprudence. At the end, the semantic search routine was experimented with the ontology populated by 15,000 judgments, all extracted from the actual jurisprudence base of the Court of Paraná, a state of Brazil. The results obtained in the experiments demonstrated that the approach was satisfactory in both the indexing of documents as well as in the semantic search, showing that the developed ontology meets the application requirements.