Refinamento de especificações de requisito pela detecção de inconsistências, ambiguidades e incompletudes através do processamento de linguagem natural

• Autor(a) principal: Mello, Otávio da Cruz
• Data de Publicação: 2024
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Manancial - Repositório Digital da UFSM
• dARK ID: ark:/26339/0013000013jtj
• Texto Completo: http://repositorio.ufsm.br/handle/1/32277
• topic: Requisitos de software; Processamento de linguagem natural; Inteligência artificial; Engenharia de requisitos; Ambiguidade; Inconsistência; Incompletude; Software requirements; Natural language processing; Artificial intelligence; Requirements engineering; Ambiguity; Inconsistency; Incompleteness; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

Context: The poor definition of requirements is a recurring issue in software projects, resulting in inconsistent, ambiguous, and incomplete specifications. These problems contribute to uncertainty in development, leading to excessive costs and schedule delays. Faced with this challenge, researchers seek approaches capable of anticipating and mitigating these issues throughout the project lifecycle, in order to define clear stories that meet user needs. Goal: The main goal of this research is proposing a technique based on Natural Language Processing to help analysts detect problematic requirements. NLP is ideal for handling requirements, as it focuses on natural language, the common form of expressing requirements. The proposed technique aims to reduce the risks associated with rework, by providing support to analysts in correcting requirement defects before they become problems for the project. Methodology: Initially, we conducted a systematic literature review to define the main problems in the Requirements Engineering area. Based on these problems, we selected challenges for a more detailed analysis, namely ambiguity, inconsistency, and incompleteness of requirements. Given that user stories are commonly expressed in natural language, we developed a solution based on Natural Language Processing, accompanied by a supporting tool for practical application of the technique. Finally, based on the application of the tool using real data from software companies, we evaluated the results obtained regarding performance metrics. Results: Through the results of the application using user stories from software companies, we noted that the system has satisfactory accuracy in detecting requirements with problems, with an average accuracy rate of about 89%. Thus, the proposed NLP-based system is a potential ally in the stages of specification and validation of requirements.