Transforming Software Requirements Classification with Deep Transfer Learning

In the dynamic landscape of software development, the classification of software requirements into functional requirements (FRs) and non-functional requirements (NFRs) is a fundamental task that impacts the entire development process. FRs define what the software should do, such as specific functionalities, while NFRs capture qualities like performance, security, and usability. Yet, the manual classification of these requirements, especially in large projects, can be a daunting challenge. Enter deep transfer learning — a solution that leverages pre-trained models to reduce training costs and improve classification accuracy, even with limited data. Paper

Why Classification of Requirements Matters

In the world of software engineering, requirements engineering is crucial. It’s estimated that around 80% of customer dissatisfaction stems from issues in the requirements phase. Efficient classification of requirements enables teams to prioritize development tasks, streamline workflows, and ultimately deliver software that meets user expectations. However, traditional approaches to classification struggle due to data scarcity, natural language complexity, and high computational costs.

Introducing Deep Transfer Learning to the Process

This study explores a hierarchical transfer learning (HTL) model that combines the strengths of natural language…