MLCQProjects: Towards evolving data set of industry-relevant software projects

Short: This data set contains three snapshots from an evolving software project data set and a script that generates those snapshots. Projects in snapshots are annotated with features used to assess industry-relevance - documentation links, installation means, support channels and whether the project delivers functionalities or samples. Context: Researchers involved in mining software repositories face a challenge that many of existing data sets reflect how software was developed in the past (typically many years ago), instead of in the present. Another challenge is that data sets based on open-source software projects include projects that are not necessarily industry-relevant. Aim: The aim of this paper is to address the aforementioned challenges and provide both: 1) a snapshot-based evolving data set of software projects (thus reflecting their present, as well as previous states), manually enriched with data considered important for industrial relevance assessment, and 2) a method of assessing industrial relevance of software projects. Method: We present a systematic method of selecting data sets of software projects for the purposes of mining software repositories of potentially industry-relevant projects and a semi-systematic method of assessing the industrial relevance of those projects. Data set: The data set contains three snapshots (spanning over 10 months) of popular Java projects from GitHub, manually enriched with industrial relevance and maintenance-related information. The presented acquisition method is sufficient to generate further snapshots and one should be able to assess the industrial relevance of the projects using the presented assessment method. Provided data set open directions of further research, e.g, 1) evaluation of code smells or defect prediction models on industry-relevant software projects prepared by independent authors and not used to build models, 2) analysis of some social aspects of open source projects (such as tooling used for providing support) and how they evolve in time.

简述：本数据集包含某演化软件项目数据集的三份快照，以及一份用于生成这些快照的脚本。快照中的项目均已通过用于评估行业相关性的特征进行标注，这些特征包括文档链接、安装方式、支持渠道，以及项目是否提供功能或示例。
背景：研究人员在开展软件仓库挖掘工作时面临两大挑战：其一，现有多数数据集仅反映了软件过往的开发状态（通常为多年前），而非当下的实际情况；其二，基于开源软件项目的数据集往往包含未必具备行业相关性的项目。
研究目标：本文旨在解决上述两类挑战，并同时提供两项成果：1） 一套基于快照的软件项目演化数据集（可同时反映项目当前与历史状态），该数据集已人工补充了对行业相关性评估至关重要的信息；2） 一种软件项目行业相关性的评估方法。
研究方法：本文提出了一种系统化的数据集遴选方法，用于筛选具备潜在行业相关性的软件项目以开展软件仓库挖掘工作，同时提出了一种半系统化的项目行业相关性评估方法。
数据集详情：本数据集包含GitHub平台上热门Java项目的三份快照（时间跨度超过10个月），并已人工补充行业相关性与项目维护相关的信息。所提出的采集方法可用于生成更多快照，且研究人员可借助本文提出的评估方法完成项目的行业相关性判定。本数据集为后续研究提供了若干开放方向，例如：1） 在独立作者构建的、未用于模型训练的行业相关软件项目上评估代码异味（code smells）或缺陷预测模型；2） 分析开源项目的部分社会属性（如用于提供支持的工具链）及其随时间的演化规律。