Predição estruturada aplicada à detecção de estrutura retórica

• Ano de defesa: 2017
• Autor(a) principal: Andreani, Alexandre Cassimiro
• 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: Estrutura retórica; Textos científicos; Predição estruturada; Algoritmos; AZPort; Brasil.; Rhetorical structure; Structured prediction; Brazil.; Ciências Exatas e da Terra; Ciência da Computação
• Link de acesso: http://repositorio.uem.br:8080/jspui/handle/1/2547

The rhetorical structure reveals whether the text is adequate with a structure established and recognized by other readers of a given literary genre. This work investigates if features coming from the structure, that is, the sequence in which the elements appear in the text contributes to its correct detection. For this, a study of structured prediction algorithms applied to the rhetorical structure detection problem was done. The objective was to discover a structured prediction algorithm that is more suitable for this type of problem. To evaluate the proposed system were used two corpora of scientific abstracts written in Portuguese and already annotated with information on the rhetorical structure. Both are composed of scientific abstracts extracted from works in several areas of Computer Science and annotated according to a rhetorical structure model composed of six categories, namely: Context, Gap, Purpose, Methodology, Result and Conclusion.In this work, this same model was used, allowing the direct comparison of the proposed predictor with the AZPort classifier. Detection using CRF, a structured prediction, had a 68% vs. 61% F1-score of the AZPort. Thus, it was verified that structured prediction algorithms benefit the task of automatically detecting the rhetorical structure of scientific abstracts.