Improving the consistency and speed of qualitative data analysis to support software engineering researchers and requirements engineering practitioners

点击翻译按钮获取中文摘要

The systematic analysis of textual data through qualitative data analysis (QDA) is used by researchers addressing questions involving real-world phenomena in context. The technique has also been proposed for industry applications, where consistency, completeness and traceability of the analysis are required. For example, in requirements engineering (RE), QDA has been proposed as a way of systematically generating specifications which are complete and which don't rely on hidden, implicit knowledge possessed only by the analyst. However, because QDA is designed to thoroughly explore the problem under consideration, it results in an unwieldy number of concepts. This means that analysts must adopt a number of techniques to ensure consistency, which are time-consuming and may still fail to ensure consistency within a single analysis, or between group members working on the same analysis. QDA is time-consuming, which affects industry use. Consistency and time-cost both severely limit adoption of QDA. Software engineering researchers have explored extending computer-assisted qualitative data analysis software (CAQDAS) by using machine learning to propose annotations or categories for later analysis, based on initial analysis. However, none of these approaches has gone beyond the domain of software engineering, or achieved results which would lead to adoption of the technique. I propose to tackle consistency and time cost through the assistive use of natural language processing. The proposal focuses on two audiences: qualitative researchers and practitioners working in requirements engineering. RE is one of the fields where QDA has been proposed as a means of ensuring completeness, consistency, and pre-requirements specification traceability. The proposal includes exploratory research into other limitations and barriers for both academic and practitioner audiences. Ultimately, the usefulness of the proposed solution is evaluated in terms of the quality of the analyses, and the experiences of the users. I expect the results to exceed the limits of the exploratory studies due to an iterative process with extensive stakeholder engagement. This is part of a broader program to identify and address limitations in CAQDAS systems, and barriers to the use of QDA by industry. Beyond the scope of this proposal, I plan to seek solutions to other problems preventing QDA use, and examine the solutions in other areas of software engineering which would benefit from the rigor of QDA. Industry funding will be sought once initial results have demonstrated the usefulness of the approach. The proposed research will impact qualitative researchers and industry in Canada and beyond, by reducing the problems associated with using QDA. The research will be supported by two PhD students, two MScs student, and one BSc students. Due to the practical application of the research, HQP will be able to acquire both research skills and industry experience.