AIDev dataset
收藏资源简介:
AIDev数据集是由魁北克大学高等技术学院研究团队构建的大规模开源软件工程数据集,专注于分析AI编码代理在GitHub项目中的贡献行为。该数据集涵盖2,807个开源项目,共包含33,596条经过人工筛选的拉取请求记录,数据来源涵盖Copilot、Devin、Cursor和Claude等主流AI编码代理生成的代码修改。数据集的构建过程通过系统化采集GitHub平台上的代理生成请求,并采用严格的过滤标准确保数据质量。该数据集主要应用于软件工程与人工智能交叉领域的研究,旨在深入理解AI代理生成代码修复的拒绝模式,为优化AI编码代理的工作流程、提升人机协作效率提供实证基础。
The AIDev dataset is a large-scale open-source software engineering dataset developed by the research team at the École de technologie supérieure, Université du Québec, with a core focus on analyzing the contribution behaviors of AI coding agents within GitHub projects. This dataset encompasses 2,807 open-source projects and contains 33,596 manually curated pull request records, with data derived from code modifications generated by leading AI coding agents including Copilot, Devin, Cursor, and Claude. The construction of the dataset involved systematic collection of agent-generated pull requests from the GitHub platform, paired with strict filtering standards to guarantee data quality. Primarily deployed for research at the intersection of software engineering and artificial intelligence, this dataset aims to uncover in-depth insights into the rejection patterns of AI-agent-generated code fixes, thereby providing an empirical foundation for optimizing the workflows of AI coding agents and enhancing human-AI collaboration efficiency.
- 1Understanding the Rejection of Fixes Generated by Agentic Pull Requests -- Insights from the AIDev Dataset魁北克大学高等技术学院 · 2026年




