Análise automática de coerência semântica em recursos acadêmicos escritos em português

• Ano de defesa: 2011
• Autor(a) principal: Souza, Vinícius Mourão Alves de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual de Maringá; Brasil; Departamento de Informática; Programa de Pós-Graduação em Ciência da Computação; UEM; Maringá, PR; Centro de Tecnologia
• Palavras-chave em Português: Resumo acadêmico; Ciência da computação; Coerência semântica; Análise automática; Análise de semântica Latente; Latent Semantic Analysis; SciPo (Scientific Portuguese); Ambiente de auxílio à escrita; Natural Language Processing.; Academic abstract; Semantic coherece; Automatic analysis of coherence; Latent Semantic Analysis.; Ciências Exatas e da Terra; Ciência da Computação
• Link de acesso: http://repositorio.uem.br:8080/jspui/handle/1/2540

The abstract can be considered one of the most important sections of an academic work. Along with the title, it is used by researchers to disseminate their research in scientic circles. In this context, Feltrim (2004) proposed an environment to support the writing of Abstract and Introduction sections called SciPo. This environment provides writing support by means of criticism and suggestions presented to the user with respect to the rhetorical (or schematic) structure identified in text submitted for analysis. Although the SciPo provides feedback indicating which parts of the text should be improved, it does not analyze features related to semantics, such as coherence, which is essential to the readability and interpretability of the text. Therefore, the main goal of this research was to develop computional resources to the automatic detection of semantic aspects of the Abstract section. We use these resources for the return of new suggestions related to coherence in the SciPo enviroment. In particular, we develop classifiers based on a set of features extracted automatically from the surface of the text and from the LSA technique - Latent Semantic Analysis and machine learning algorithms. Thus, the classifiers provide indications on the semantic aspects that contribute to the abstract is considered coherent. Both the intrinsic assessments of the classifiers as the evaluation of the prototype in a context of use with real users demonstrated the potential of classifiers to aid writing academics abstracts with higher lever of coherence through new suggestions.