ldentificação de especialistas em APIs a partir de conhecimento existente em repositórios sociais de software

• Ano de defesa: 2018
• Autor(a) principal: Macedo, Camille Nogueira de
• Orientador(a): Bittencourt, Roberto Almeida
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual de Feira de Santana
• Programa de Pós-Graduação: Mestrado em Computação Aplicada
• Departamento: DEPARTAMENTO DE CIÊNCIAS EXATAS
• País: Brasil
• Palavras-chave em Português: Identificação; Especialista; API de software; Métricas
• Palavras-chave em Inglês: Identification; Expert; Software API; Metrics
• Área do conhecimento CNPq: METODOLOGIA E TECNICAS DA COMPUTACAO::ENGENHARIA DE SOFTWARE
• Link de acesso: http://tede2.uefs.br:8080/handle/tede/776

Identification of software development experts usually represents high operational costs for companies. In order to mitigate this problem, some researchers have presented different strategies to find experts. Despite their efforts, such strategies pointo to particular solutions and particular evaluations, leading to different conclusions despite their use of similar inputs. In this work, we built an understanding of the field by selecting some of the recent metrics, and proposed a tool that can use these metrics to identify experts in software APIs from their use of the source code in a set of projects available on a social software repository. From these results, we performed an exploratory study with three software APIs with the goal of evaluating five metrics to identify experts from source code. In this evaluation, we produced expert rankings from the metrics computed by the tool, and we perceived that these metrics show a strong correlation with each other. We also evaluated the metrics against a ground truth based on software development performed after computing the metrics. Results show that, for a small scenario of developers who use the API and with less API complexity, the metrics show better precision. For larger developer groups and larger API complexity, results are less accurate, but still have an average accuracy of 48\% from Top-5 rankings and for the three APIs used in the evaluation. This work adds to the body of knowledge on automatic determination of software expertise, pointing to the feasibility of the techniques, and presenting an evaluation of the potential use of expertise metrics in the context of software APIs used in social software repositories.