arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2311.03358cs.SE
收藏

采用知识图谱方法理解与分析提交消息中的设计 rationale

Towards Understanding and Analyzing Rationale in Commit Messages using a Knowledge Graph Approach

Mouna Dhaouadi, Bentley James Oakes, Michalis Famelis

AI总结:

本研究构建Kantara端到端设计rationale重构流水线,通过知识图谱结构化提交消息中的rationale信息并提取分类,同时构建Linux内核OOM组件标注数据集,评估显示XGBoost等NLP分类技术效果良好。

摘要:

从提交消息中提取设计 rationale 信息可帮助开发者更好地理解系统及其过往开发历程。本文介绍了我们正在开展的 Kantara 端到端设计 rationale 重构流水线工作,该流水线包含三部分:a) 将设计 rationale 信息构建为基于本体论的知识图谱结构;b) 从提交记录中提取并分类这类信息;c) 为开发者生成分析报告与可视化结果。我们还介绍了为 Linux 内核的 Out-of-Memory 组件运行示例构建标注数据集的工作,该数据集被用作评估 NLP 分类技术的基准真值,评估结果显示相关技术效果良好,尤其是多分类技术 XGBoost。

英文摘要:

Extracting rationale information from commit messages allows developers to better understand a system and its past development. Here we present our ongoing work on the Kantara end-to-end rationale reconstruction pipeline to a) structure rationale information in an ontologically-based knowledge graph, b) extract and classify this information from commits, and c) produce analysis reports and visualizations for developers. We also present our work on creating a labelled dataset for our running example of the Out-of-Memory component of the Linux kernel. This dataset is used as ground truth for our evaluation of NLP classification techniques which show promising results, especially the multi-classification technique XGBoost.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑