SHF: Small: Automated Fine-Grained Requirements Traceability

点击翻译按钮获取中文摘要

Requirements traceability is mandatory in safety-critical domains (such as aerospace, transportation, communication, medical device software, etc.) and it is essential in supporting acceptance testing or regulatory reviews. The creation and maintenance of traceability links between requirements and code has been identified by researchers as one of the grand challenges in software traceability. Given that most requirements are written in natural language and source code also contains large amounts of natural language, researchers proposed using text-retrieval techniques for automated traceability-link recovery. Text-retrieval-based approaches inherently suffer from low accuracy due to the vocabulary mismatch between requirements and source code and the coarse-grained granularity of the retrieved code (such as classes, files, methods). As a consequence, developers rarely use traceability links in everyday tasks related to code change. Improving the automated recovery between requirements and code will help with important daily tasks faced by the software engineers (such as bug localization and code reviews), which in turn will have a potentially large economic impact, as it may lead to higher quality software and less development effort. In addition, regulatory reviewers of safety-critical software are expected to adopt results of this research project to improve their work. The impact of their work on society is expected to be high as a result. This project redefines the requirement-to-code traceability-retrieval problem into a heuristic-driven approach, using static code analysis and text analysis, which accurately matches functional constraints embedded in requirements to their source code implementations. One contribution of the project will be the identification and classification of functional constraints present in requirements, written in natural language. Central to the project is the hypothesis, supported by preliminary research, that developers use well-defined patterns to express and implement the functional constraints in code, which will be discovered and cataloged. The functional constraints described in requirements will be automatically identified using natural-language-processing techniques and deep learning. The implementation patterns will be automatically identified using precise static-analysis techniques. Overall, the novel requirements-to-code traceability paradigm will allow recovering links at various granularity levels, as needed by various users. Finally, the new type of traceability links will be used to improve essential software-development tasks, namely bug localization and code reviews.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.