Capturing creative requirements via requirements reuse: A machine learning-based approach

Abstract

The software industry has become increasingly competitive as we see multiple software serving the same domain and striving for customers. To that end, modern software needs to provide creative features to improve sustainability. To advance software creativity, research has proposed several techniques, including multi-day workshops involving experienced requirements analysts, and semi-automated tools to support creative thinking in a limited scope. Such approaches are either useful only for software with already rich issue tracking systems, or require substantial engagement from analysts with creative minds. In a recent work, we have demonstrated a novel framework that is beneficial for both novel and existing software and allows end-to-end automation promoting creativity. The framework reuses requirements from similar software freely available online, utilizes advanced natural language processing and machine learning techniques, and leverages the concept of requirement boilerplate to generate candidate creative requirements. An application of our framework on software domains: Antivirus, Web Browser, and File Sharing followed by a human subject evaluation have shown promising results. In this invited extension, we present further analysis for our research questions and report an additional evaluation by human subjects. The results exhibit the framework’s ability in generating creative features even for a relatively matured application domain, such as Web Browser, and provoking creative thinking among developers irrespective of their experience levels.