Quality Assurance of System Requirements Based on Large Language Models

Software requirements play a critical role in safety-critical cyber-physical systems as they are the foundation for designing, implementing, and testing of the software systems. Due to limited budget and time constraints, software requirements in a commercial setting tend to have unsatisfactory quality, which complicates downstream development activities. Prior research identifies dozens of quality attributes of software requirements such as completeness, correctness, and consistency. However, quality attributes can vary across different domains, especially within the automotive system. Currently, there remains a dearth of research in this field. Moreover, large language models have the potential to identify and remediate issues with respect to different quality attributes. We plan to identify a prioritized list of quality attributes by working with our industry partners. Next, to incorporate domain knowledge into large language models, prompting and fine-tuning are the most popular techniques. Finally, large language models will be used to identify and remediate quality issues in production-level software requirements. In summary, the goal of this project is to apply large language models on software requirements for commercial safety-critical cyber-physical systems to identify potential issues and provide fixes.

Faculty Supervisor:

Lionel Briand

Student:

Partner:

Wind River International Ltd.

Discipline:

Computer science

Sector:

Information and cultural industries

University:

University of Ottawa

Program: