需求跟踪行为认知与主动学习机理研究

Ming the apriori knowledge of traceability from the massive tracing trajectory and relevance information between artifacts, and providing the traceability navigation services for developers with intelligent and friendly interface, is a hot but difficult topic in the fields of software maintenance and requirement engineering. With the construction of the human-centred traceability process model , this project will focus on the multi-dimensional artefact correlation model, the tracing behaviors representation model and the rule-learning model, to explore the tracing behaviors cognition and its active learning mechanism based on massive tracing trajectory and relevance information between artifacts. In particular, firstly, a human-centred traceability process model will be constructed to define the tracing life-cycle and build a body of knowledge that reflects best practices. Then, an artifact correlation model will be proposed to fuse the correlation between artefacts from different dimensions to characterize the macroscopic and microscopic feature of tracing behaviors. After that, a rule-learning model will be presented for the following three purposes: 1) infers developer’s intention and then recommends candidate traceability link based on his or her tracing trajectory and preference information; 2) optimizes itself according to the feedback such as the developer’s selection of recommendations; 3) meets the tracing behavior apriori knowledge extraction and active learning demand. Finally, based on the work outlined above, this project will further develop corresponding tracing tools, and verify the effectiveness of these methods and tools by several representative open source projects.

如何从海量制品关联信息和跟踪轨迹数据中有效挖掘需求跟踪先验知识，进而为开发人员提供智能友好的主动需求跟踪导航服务一直是国内外在软件维护和需求工程等方面的研究热点和难题。本课题拟构建以人为中心的需求跟踪过程模型，并重点研究多维度制品相关性度量模型、跟踪行为表征模型以及规则学习模型，以探索基于海量制品关联信息和跟踪轨迹数据的跟踪行为认知和主动学习机理。其中，需求跟踪过程模型将定义需求跟踪生命周期，并承载体现最佳跟踪实践的知识体系；制品相关性度量模型将融合不同维度的制品关联信息，以表征跟踪行为的宏观和微观特征；而规则学习模型将基于开发人员的跟踪轨迹和偏好信息推测其探索意图并推荐候选跟踪关系，再根据开发人员对推荐结果的选取等反馈信息进行自我优化，并提供跟踪行为先验知识提炼及其主动学习机制。在上述研究工作的基础上，课题还将开发相应的需求跟踪工具，并通过具有代表性的开源项目验证相关方法和工具的有效性。

尽管需求跟踪的重要性已被广泛认可，并且也取得了大量的研究成果，但是，研究人员至今仍然对开发人员的跟踪行为及其先验知识知之甚少。针对这一问题，课题重点研究了需求跟踪行为认知与主动学习机理，其主要内容包括：构建了以人为中心的需求跟踪过程模型，以获取和管理需求跟踪的过程信息和知识；研究了适用于多种软件制品类型，且综合多维度制品关联信息的制品相关性度量模型，以增强需求跟踪工具的导航能力；研究了一种交互迭代式需求跟踪方法，该方法能够主动学习跟踪行为所蕴含的先验知识，基于开发人员对跟踪关系的选取等反馈信息进行自我优化；研究了基于自然语言处理技术的需求跟踪方法，以减轻制品中的噪音对需求跟踪带来的不利影响；将需求跟踪转换为更为一般的本体匹配问题，为需求跟踪行为认知和主动学习机理研究提供了新的视角。基于以上研究结果，能够有效提炼出在海量制品关联和跟踪轨迹中的需求跟踪行为知识，提高需求跟踪工具对上下文的感知能力以及在各种复杂场景中的探索能力,以提供更为智能友好的需求跟踪导航服务。