基于复合网络挖掘的软件开发团队智能理解、预测与优化研究

Software systems are usually developed by software developers in the form of teams. In addition to the individual intelligence, team intelligence plays important roles for the productivity of software development. Therefore, understanding the team intelligence in software development, predicting the performance of team intelligence, and optimizing the teams for better team intelligence are key research paths to improve software productivity. However, comprehensive understanding, prediction, and optimization of team intelligence are challenging due to the composite network nature of software development. That is, software development can be seen as the composite of software developers and their relationships, software development tasks and their relationships, and software development skills and their relationships. To address the composite network challenge, this project plans to systematically study several key research problems including software development team intelligence understanding via network fusion and network embedding, software development team intelligence prediction via nonlinear modeling and cross-domain learning, and software development team intelligent optimization via multi-objective optimization and graph matching.

软件系统通常由多个软件开发者以团队的形式共同完成开发任务。影响这类软件开发效率的因素除了软件开发者的个体智能，还包括这些个体智能组合起来的团队智能。因此，理解软件开发的团队智能，预测团队智能的效果，并从团队成员组合的角度优化团队智能是提高软件开发效率的重要手段。然而，全面合理地理解、预测和优化软件开发团队智能还存在一些挑战，其根源在于软件开发的复合网络特性，即软件开发是由软件开发任务之间的关系网络、软件开发者之间的关系网络、软件开发技能之间的关系网络所复合而成的网络共同作用的过程。鉴于此，本项目拟从软件开发的复合网络特性出发，系统研究基于网络融合与网络嵌入的软件开发团队智能理解、基于非线性建模与跨领域学习的软件开发团队智能预测、基于多目标优化与图匹配的软件开发团队智能优化等关键科学问题，从而为高效率的软件开发提供支持。

软件开发活动作为一种团队协作活动，理解其团队协作的机理显得尤为重要。本项目针对软件开发活动的复合网络特性，系统研究了基于网络融合与网络嵌入的软件开发团队智能理解、基于非线性建模与跨领域学习的软件开发团队智能预测、基于多目标优化与图匹配的软件开发团队智能优化等关键科学问题，取得显著进展。这些研究工作为更好地理解软件开发活动提供了支撑与依据，并且通过在缺陷分配、缺陷定位、软件团队成员推荐等任务中的实证研究中验证了效果。相关研究内容共发表论文12篇，其中CCF A类3篇，CCF B类5篇，申请专利2项，并产出了一系列原型系统。