Construção automática de resumos gráficos utilizando processamento de linguagem natural

• Ano de defesa: 2018
• Autor(a) principal: Santos, Vinicius dos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Cornelio Procopio; Brasil; Programa de Pós-Graduação em Informática; UTFPR
• Palavras-chave em Português: Mapeamento conceitual; Processamento de linguagem natural (Computação); Resumos; Concept mapping; Natural language processing (Computer science); Abstracts; CNPQ::CIENCIAS EXATAS E DA TERRA; Ciência da Computação
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/3282

Context: Secondary studies, such as Systematic Literature Reviews (SLR) and Systematic Mappings (SM), have been increasingly used in Software Engineering (SE) since they allow the identification of available evidence related to a research topic. One of the main activities of the process of conducting a secondary study is the primary studies selection, which involves, at first, the reading of the abstracts of the candidate studies. However, with the growing number of scientific publications, coupled with the poor quality of their abstracts, it makes this activity increasingly difficult for researchers. Some solutions have been proposed to mitigate the problem, among them, the use of structured abstracts and graphic summaries. Previous studies have proposed guidelines for the construction of graphic summaries. However, these summaries continue to be created manually. Objectives: This work has two objectives: (i) understand the use of Conceptual Maps (CM) in Computer Science and to investigate the main techniques for generation of MCs from Natural Language Processing (NPL); (ii) propose an approach for the automatic construction of graphic abstracts based on CMs using NLP techniques. Method: initially the collection of the main practices for the construction of CMs from NLP was performed. Next, an approach for the construction of graphic summaries based on CMs was defined. Finally, evaluations were conducted in order to verify the quality of the CMs generated. Results: The pilot experiment conducted showed that the CMs constructed by the initiative demonstrated a good performance in terms of concept extraction and comprehensiveness when representing the concepts of the abstract. Conclusions: The preliminary results show that the proposed initiative can generate valid propositions and represent graphic summaries through CMs, becoming an important tool to summarize a complex structure of textual information, contributing to the identification of the most important information of an article.