面向开源生态的大众贡献智能化汇聚机理与方法研究

Nowadays, a large number of crowd developers join the open source ecosystem to achieve software creation and production, which makes a significant impact on software innovation. However, the main driven factors of software innovation are not only originated from the inspirations from crowds, but also depend on a highly effective process that aggregating and transforming crowd contributions to high-quality and reusable software artifacts. Because of the high volume of incoming contributions from crowds, there is in urgent need of exploring an effective and efficient aggregation mechanism to achieve this goal. In this project, we propose a new research problem focusing on the intelligent aggregation theories and mechanisms based on the mining the big data from software life-cycle, and design a multi-dimensional framework of quantitative measurement. Then, using the data-driven and knowledge-driven approach, we can refactor the crowd contributions and link them to the actual development tasks, continuously evaluate the quality, and crowdsourcing review tasks to suitable reviewers. Furthermore, an intelligent aggregation system of human-machine cooperation can be constructed based on the above techniques. This project will provide new software theories and technical tools for crowd-based software innovation.

当前围绕开源生态展开的大众化软件创作与生产活动，已成为软件持续创新的源动力。然而，软件创新不仅需要充分激发大众群体的创作灵感与贡献热情，还应持续汇聚收敛大众的碎片化创意，使其凝结为高质量、可复用的软件制品。因此，面对大规模持续涌入的大众贡献，如何高效地实现大众贡献汇聚，是当前开源生态软件创新亟待解决的难题。本项目立足于软件全生命期大数据，提出大众贡献的智能化汇聚这一全新问题，构造面向贡献汇聚的多元量化分析理论框架，将数据驱动和知识驱动相融合，实现创意与任务的重构与映射、代码贡献质量的持续性评估、众包审查任务的生成与分派，进而形成一种人机高效协同的智能化汇聚体系，为加速大众化软件创新提供新的技术理论和支撑工具。

开源运动经过数十年的蓬勃发展，已经在全球性软件工程领域占据了主导地位。围绕开源生态展开的软件创作与生产活动，已成为软件持续创新的重要动力。然而，大众化开源贡献具有数量急剧增长、类型多样、质量参差不齐、贡献者能力水平发展不均衡等特点。如何持续高效地实现开源贡献汇聚，已经成为当前开源生态软件创新亟待解决的全新问题。本项目立足于开源生态大数据，提出大众贡献的智能化汇聚这一全新问题，其具体内涵为：当大规模群体的个性化创意被充分激发与释放后，以开源生态中潜在的软件知识为驱动，形成一种人机高效协同的智能化贡献汇聚体系，为加速开源生态协同质量和效率提供理论指导和方法支撑。围绕上述研究问题，本项目聚焦贡献汇聚机理量化度量理论、软件任务重构与管理方法、贡献质量持续性评估与群体行为感知四个方面开展研究。本项目构建了群体自主协同场景下的典型数据集，提出了数据标注及分析框架，深入分析了群体协同实践模型，实现了上下文启发式任务摘要生成、自动化缺陷预测、引导化缺陷修复、层次化代码标注、以及历史关联性行为感知等技术突破，为大规模开源贡献汇聚质量与效率提升提供理论与工具支持。