MaRV: A Manually Validated Refactoring Dataset

Despite the existence of traditional refactoring tools that offer semi-automated assistance, machine learning-based models have shown significant potential to generate refactored code. A comprehensive, manually validated refactoring dataset could help the software engineering community to train such models for effective refactorings. However, the community lacks a manually validated refactoring dataset. This paper introduces the MaRV dataset containing 693 manually evaluated code pairs extracted out of 126 GitHub Java repositories, representing four types of refactoring. In addition, the metadata describing the supposedly refactored elements was collected. Each code pair was manually evaluated by two reviewers out of 40 participants. MaRV dataset is constantly evolving with a web-based tool available for evaluating refactoring representations. The potential application of this dataset is to improve the accuracy and reliability of state-of-the-art models in refactoring tasks (e.g., refactoring candidate identification and refactoring code generation) by providing high-quality data.