Developers assignment for analyzing pull requests
Manoel Limeira de Lima Júnior, Daricélio Moreira Soares, Alexandre Plastino, Leonardo Gresta Paulino Murta · 2015
A new collaboration approach is becoming increasingly common in open-source projects: the pull request model. In this kind of collaboration, developers that do not belong to the core team of a project can submit contributions to the core team. In projects that receive many pull requests, the task of assigning developers to analyze them is a difficult one. In this work, we propose to use data mining techniques, more specifically, classification strategies, in order to suggest the most appropriate developers to analyze a contribution, considering the pull request model. The experiments were conducted using 21 open source projects, each one characterized by 14 attributes. The first set of experiments aimed at indicating just one developer to analyze the pull request. The obtained predictive accuracy ranged from 22.45% to 68.27%. The Random Forest classifier achieved the best result in 76% on the projects. In the second set of experiments, we conclude that, when suggesting three developers to analyze a pull request, the chance of identifying the developer that actually analyzed the pull request ranged from 47.33% to 95.47%.