Who Should Review This Change? Putting Text and File Location Analyses Together for More Accurate Recommendations

• Main Authors: XIA, Xin, David LO, WANG, Xinyu, YANG, Xiaohu
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2015
• Subjects: Modern Code Review; Path Similarity; Recommendation System; Text Mining; Software Engineering
• Online Access: https://ink.library.smu.edu.sg/sis_research/3086; http://dx.doi.org/10.1109/ICSM.2015.7332472
• Institution: Singapore Management University

Software code review is a process of developers inspecting new code changes made by others, to evaluate their quality and identify and fix defects, before integrating them to the main branch of a version control system. Modern Code Review (MCR), a lightweight and tool-based variant of conventional codereview, is widely adopted in both open source and proprietary software projects. One challenge that impacts MCR is the assignment of appropriate developers to review a code change. Considering that there could be hundreds of potential code reviewers in a software project, picking suitable reviewers is not a straightforward task. A prior study by Thongtanunam et al. showed that the difficulty in selecting suitable reviewers may delay the review process by an average of 12 days. In this paper, to address the challenge of assigning suitable reviewers to changes, we propose a hybrid and incremental approach Tie which utilizes the advantages of both Text mIning and a filE location-based approach. To do this, Tie integrates an incremental text mining model which analyzes the textual contents in a reviewrequest, and a similarity model which measures the similarity of changed file paths and reviewed filepaths. We perform a large-scale experiment on four open source projects, namely Android, OpenStack, QT, and LibreOffice, containing a total of 42,045 reviews. The experimental results show that on average Tie can achieve top-1, top-5, and top-10 accuracies, and Mean Reciprocal Rank (MRR) of 0.52, 0.79, 0.85, and 0.64 for the four projects, which improves the state-of-the-art approach RevFinder, proposed by Thongtanunam et al., by 61%, 23%, 8%, and 37%, respectively.