LabelGit

LabelGit是由格罗宁根大学伯努利研究所创建的一个用于Java软件仓库分类的数据集，包含11,502个条目。该数据集利用源代码的语义和依赖图，通过直接从源代码中提取信息，如依赖图和标识符的源代码神经表示，来进行分类。创建过程中，数据集从GitHub和Awesome-Java列表中收集项目，并通过依赖图提取、特征提取、标签重映射和数据增强等步骤进行处理。LabelGit旨在解决软件分类问题，特别是在组件化软件开发中，通过不依赖于项目代理（如README）的方法，使用整个源代码进行分类，从而推动机器学习在软件工程中的应用。

LabelGit is a dataset for Java software repository classification developed by the Bernoulli Institute of the University of Groningen, comprising 11,502 entries. This dataset leverages source code semantics and dependency graphs, performing classification by directly extracting information from source code—including dependency graphs and neural representations of source code identifiers. During its development, projects were sourced from GitHub and the Awesome-Java curated list, with the dataset processed via steps including dependency graph extraction, feature extraction, label remapping, and data augmentation. LabelGit aims to solve software classification tasks, especially in component-based software development, by employing full source code for classification without relying on project proxies such as README files, thus promoting the application of machine learning in software engineering.