Improving Code Review by Predicting Reviewers and Acceptance of Patches

Code reviews are an important part of software development because they help to increase code quality and reliability. For this paper, we observed the review processes of two open-source projects, Firefox and Mozilla Core. We noticed that code reviews are mostly organized manually. In particular, finding appropriate reviewers is a complex and time-consuming task and, surprisingly, impacts the review outcome: review requests without an initial reviewer assignment have lower chances to be accepted (and take longer). Based on our observations we propose two improvements: (1) predict whether a given patch is acceptable and (2) suggest reviewers for a patch. We implemented and tested both approaches for the Firefox and Mozilla Core projects. In our experiments, the prediction accuracy was 73% for the the review outcome and 51-80% for the reviewer recommendation. The values for accuracy are higher than those of comparable approaches and are high enough to be useful and applicable in practice.