Towards achieving effective traceability management between and within software artifacts
Modern software systems are increasingly complex, making effective traceability (e.g., linking requirements, design, code, and tests) essential yet challenging. In his thesis, Ahmad Zaki Bin Mohamad Pauzi addresses these challenges by leveraging Natural Language Processing (NLP) and data-driven methods to improve traceability across diverse software artifacts.
First, Bin Mohamad Pauzi presents a systematic mapping study of 96 papers (2013–2021), identifying key barriers such as representation similarity, scalability, and explainability in NLP-based traceability. Building on these insights, Bin Mohamad Pauzi introduces a novel meta-analysis framework combining domain-specific Word2Vec embeddings and ARIMA time series modeling to forecast emerging research trends. To enhance semantic understanding, he proposes a domain-centric approach, using embedding-based mapping to connect conceptual elements across heterogeneous artifacts. Bin Mohamad Pauzi also investigates the evolution of design artifacts, particularly UML class diagrams in long-standing open-source projects, revealing patterns of “UML dormancy” and misalignment between design and implementation.
Finally, Bin Mohamad Pauzi bridges theory and practice through a multivocal mapping study and an industrial experience report, demonstrating the successful deployment of end-to-end traceability in a real-world DevOps environment. These findings offer actionable insights for improving traceability tools and processes in industry. Overall, this thesis contributes a comprehensive, multi-dimensional framework for advancing traceability in software engineering, promoting more maintainable, transparent, and sustainable software systems.