Um estudo da representação de documentos jurídicos em espaços métricos

• Ano de defesa: 2019
• Autor(a) principal: Machado, Gustavo Menezes
• Orientador(a): Montalvão Filho, Jugurta Rosa
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Programa de Pós-Graduação: Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Computação; Jurisprudência; Multidimensional scaling; Julgamentos; Textos jurídicos
• Palavras-chave em Inglês: Multidimensional Scaling (MDS); Doc2Vec; Jurisprudence; Judgments; Judicial documents
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://ri.ufs.br/jspui/handle/riufs/14553

Dozens of decisions are made daily from interpretations of the laws made by courts across the country. This set of similar decisions on the same subject is known as jurisprudence and serves as the basis for future judgments and arguments. In legal texts written in Brazilian Portuguese, in addition to words being guided by different esthetic rules, there is also the use of frequent references to legal elements, which makes the analysis of legal texts written in Brazilian Portuguese a stimulating problem. This work explores a metric space associated with contexts and the sharing of symbols between contexts of legal documents, that is, the search for spaces suitable for the representation of texts as court lawsuits, where each process - or part of it - is represented as a point, and the distances between these points represents probabilistic measures. For such representation, the Multidimensional Scaling (MDS) was used, which is a technique of dimensionality reduction where the relations of distances between the points in the projected space approximate the proximity measurements of the objects of the original space. The case law of the Sergipe State Court of Justice was used, in addition to a controlled set of words used in the legal area, provided by the Federal Supreme Court. The experiments showed that the proposed method obtained a better classification in 43.5% of the cases, while Doc2Vec was superior in only 35.7% of the cases, evidencing the existence of a more adequate metric space for the representation of legal texts written in Brazilian Portuguese than a space purely based on co-occurrence of symbols, as found by Doc2Vec.