A hybrid approach to code reviewer recommendation with collaborative filtering

Code review is known to be of paramount importance for software quality assurance. However, finding a reviewer for certain code can be very challenging in Modern Code Review environment due to the difficulty of learning the expertise and availability of candidate reviewers. To tackle this problem, existing efforts mainly concern how to model a reviewer's expertise with the review history, and making recommendation based on how well a reviewer's expertise can meet the requirement of a review task. Nonetheless, as there are both explicit and implicit relations in data that affect whether a reviewer is suitable for a given task, merely modeling review expertise with explicit relations often fails to achieve expected recommendation accuracy. To that end, we propose a recommendation algorithm that takes implicit relations into account. Furthermore, we utilize a hybrid approach that combines latent factor models and neighborhood methods to capture implicit relations. Finally, we have conducted extensive experiments by comparing with the state-of-the-art methods using the data of 5 popular GitHub projects. The results demonstrate that our approach outperforms the comparing methods for all top-k recommendations and reaches a 15.3% precision promotion in top-1 recommendation.