Some of the most difficult challenges in software are related to requirements creativity. Building on previous work, this well-organized study investigates the extent to which the creation of requirements can be automated. Focusing on existing systems, the study leverages a mix of technologies, including social networks and network clustering, topic modeling and knowledge allocation, and combinational creativity and semantic analysis.
The paper begins with a conceptual discussion of types of creativity, network models of stakeholders and associated artifacts, and allocation topics to verb-noun pairs. It continues with a review of the authors’ prior work on a semi-automated creativity framework that utilizes those models in a five-step process. The steps identify typical topics and combine them in atypical ways, from which human analysts might generate novel and useful requirements.
The study extends that framework to a fully automated one by replacing the last step using verb-clause pairs, derived from original requirements and stakeholder communications, and associated syntactic templates to generate requirements. The study applies the frameworks to two large open-source systems and assesses the performance and quality of the generated requirements.
The paper will interest requirements, software, and system engineers; information analysts; and those who study creativity.
The difficult challenges of requirements creativity continue to allude us, but the authors’ framework helps us understand what steps can be automated, how a mix of current technologies can perform parts of the automation, and the limitations of those technologies in creating novel and useful requirements.