Enhancing requirements quality through automated ambiquity detection: A Comparative Analysis of Manual, Rule-Based and Generative Artificial Inteligence Techniques

Abstract

Ambiguity in requirements engineering specifications is a longstanding challenge in software en- gineering, often contributing to costly misunderstandings, delays, and system failures. This thesis investigates the potential of Generative Artificial Intelligence (GenAI), specifically Large Lan- guage Models (LLMs), to enhance the requirements review process by improving ambiguity de- tection and resolution. Traditional approaches to requirements review-such as manual inspections, rule-based analy- ses, and template-driven techniques-have played a crucial role in managing ambiguity. However, these methods often struggle with issues of scalability, semantic interpretation, and contextual sensitivity. This study conducts a comparative analysis of traditional review practices and GenAI- enabled approaches, assessing their respective strengths and limitations in the context of ambiguity detection. The research develops a structured evaluation framework that highlights where GenAI mod- els can offer meaningful improvements, particularly in understanding context, capturing nuanced semantics, and adapting to varied domains. Based on this analysis, a hybrid review strategy is proposed-one that combines the procedural reliability of traditional methods with the advanced analytical capabilities of GenAI. Implementation considerations are also addressed, including challenges related to model trans- parency, organizational trust, and integration with existing workflows. The findings suggest that, when applied responsibly, GenAI can significantly improve the effectiveness and efficiency of requirements review activities. This thesis contributes to the field of AI-assisted requirements engineering by offering a pro- totype implementation of a project called ReqVision that is an Agent and prompt builder to Large Language Models evaluating and adopting pre-trained model on a pilot project of an Infusion Pump in the Medical domain. It aims to improve the Requirement engineering process review on safe critical systems.