Quim Motger PhD thesis

Advisors: Xavier Franch and Jordi Marco

Abstract of the thesis

Requirements engineering (RE) is fundamental to successful software development, especially in modern, large-scale projects. Efficient management of text-based artefacts is key to accurate elicitation, refinement, and validation of requirements. Despite the industrial trend towards adopting natural language processing (NLP) methods, challenges in their pervasiveness, reliability, scalability, and reusability persist. Moreover, the advent of large language models (LLMs) has set the groundwork for further research in automated document analysis in the field of RE. This thesis explores the integration of NLP methods and tools to automate and enhance RE tasks (NLP4RE) in three document-oriented areas: requirements traceability, requirements analysis for information retrieval, and requirements feedback gathering. For requirements traceability, methods for dependency and duplicate detection in text-based requirements documents are proposed and evaluated. In requirements analysis, a knowledge graph-based approach is developed to create adaptive, crowdsourced repositories of RE-related documents. For requirements feedback gathering, techniques for extracting features from user reviews and analyzing feedback are presented and evaluated. This research is shaped in the context of multiple case and sample studies validated empirically, demonstrating their effectiveness in real-world scenarios. The contributions presented in this thesis entail advancements in streamlining RE tasks and improving the accuracy, efficiency and adoption of NLP4RE tools and methods. Ultimately, this thesis aims to provide novel insights, methodologies and technical contributions to the NLP4RE field.