Recommending tags for pull requests in GitHub

Abstract

Abstract Context In GitHub, contributors make code changes, then create and submit pull requests to projects. Tags are a simple and effective way to attach additional information to pull requests and facilitate their organization. However, little effort has been devoted to study pull requests’ tags in GitHub. Objective Our objective in this paper is to propose an approach which automatically recommends tags for pull requests in GitHub. Method We make a survey on the usage of tags in pull requests. Survey results show that tags are useful for developers to track, search or classify pull requests. But some respondents think that it is difficult to choose right tags and keep consistency of tags. 60.61% of respondents think that a tag recommendation tool is useful. In order to help developers choose tags, we propose a method FNNRec which uses feed-forward neural network to analyze titles, description, file paths and contributors. Results We evaluate the effectiveness of FNNRec on 10 projects containing 68,497 tagged pull requests. The experimental results show that on average, FNNRec outperforms approach TagDeepRec and TagMulRec by 62.985% and 24.953% in terms of F 1 − s c o r e @ 3 , respectively. Conclusion FNNRec is useful to find appropriate tags and improve tag setting process in GitHub.