Leveraging LLMs for Coverage Analysis from High to Low-Level Requirements

Abstract

The increasing capabilities of Large Language Models (LLMs) have opened new av enues for automating critical tasks in software and systems engineering, particularly in the generation and analysis of requirements. This thesis investigates the poten tial of LLMs to generate Low-Level Requirements (LLRs) from existing High-Level Requirements (HLRs), aiming to evaluate how effectively these models can replicate human-derived requirements. The study employs multiple state-of-the-art LLMs to generateLLRsusingvariousprompting strategies,followedbyasystematicevaluation of coverage and traceability through expert validation and alignment with established academic criteria. The findings of this research provide insights into the strengths and limitations of LLMs in capturing detailed, actionable system specifications. While LLMs demon strate significant potential for automating certain aspects of requirements engineer ing, humanexpertiseremainsessentialforensuringcompleteness, accuracy, andcon textual relevance. By highlighting areas where automation can enhance efficiency and where human oversight is necessary, this work contributes to a deeper under standing of LLM capabilities in requirements engineering and offers practical recom mendations for integratingthesemodels intoprofessionalpracticetoimproverequire ment traceability, consistency, and quality.