On the Classification of Refactoring Code Reviews

Code review has become standard practice in both business and open source projects with the objective of improving software quality, promoting knowledge exchange, and ensuring adherence to coding rules and guidelines. It involves developer discussions on potential refactoring operations before incorporating changes into the code base. Despite extensive investigations into the general challenges, best practices, and socio-technical elements of code review, there is limited understanding about the assessment of refactoring and what developers prioritize when examining refactored code. Therefore, our study aims to comprehend the primary factors developers consider when reviewing refactored code. To do so, we design a model that takes as input a given code review conversation, and classifies it into one refactoring category, e.g., quality, integration, testing, etc. Multiple machine learning approaches are used and evaluated, and the experimental results are examined qualitatively and quantitatively at various granularities. Refactoring-related code review category prediction is found to be possible with high consistency, achieving an average scores of 0.8 on accepted universal classification metrics, through the utilization of contemporary and traditional machine learning approaches. These findings suggest that refactoring code review categorization can be embedded in traditional workflows to increase efficiency and decrease the time to production in software development.