A influência dos parâmetros de Análise por Semântica Latente aplicada a localização de defeitos de software

• Ano de defesa: 2011
• Autor(a) principal: Silva, Allysson Costa e
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação em Ciência da Computação; Ciências Exatas e da Terra; UFU
• Palavras-chave em Português: Defeitos de software; Recuperação de informação; Parâmetros de busca; Código fonte; Indexação por semântica latente; Software; Software - Desenvolvimento; Bugs; Information retrieval; Search parameters; Source code; Latent semantic indexing; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/12505

During its life cycle, software systems must pass through continuous change to have bugs fixed and get a reasonable conformance between user requirements and implemented software functions. The necessary effort to execute such changes that occur in the software maintenance phase is considerable and is influenced by the developer software comprehension level. In this way, the production and improvement of tools related to software comprehension can speed up and minimize software maintenance efforts. In this work, a tool to retrieve the traceability links between documentation (bug descriptions) and source code was implemented using an Information Retrieval technique. The main purpose of this work is the analysis of the influence of Latent Semantic Indexing (LSI) parameters values on accuracy and performance of the tool. The accuracy was measured with the number of methods that should be checked by the developer to find a bug. This study has shown that the parameter s values have direct impact on software maintenance accuracy and performance. The predominant parameter s values were: dimensionality reduction k=300; positive use of weighting functions for method and class names and for method and class name fragments; use of ET or QTL source code filters. Additionally, it was possible to identify semantic alignments between the user vocabulary and method and class names.