Enhancing Requirements Change Impact Analysis in Critical Systems Using Natural Language Processing

Abstract

Machine learning has significantly impacted various domains by extracting meaningful patterns from large datasets, yet its integration into requirements engineering (RE) for safety-critical sys- tems (SCS) presents unique challenges. This study offers a overview of the current research land- scape and identifies promising directions for future work in the application Natural Language Processing (NLP) techniques to enhance RE tasks in the context of SCS. Starting by framing the interrelated nature of SCS and the normative regulation, the research identifies untapped potential for usage of NLP methods. Particularly in case of SCS, we see ef- ficiency potential in the development of tools for assistance in estimation effort and safety risks involved by introducing a change in SCS. The goals of this research are categorized into three specific activities: 1. Supporting Change-Impact Analysis (CIA). 2. Tracing or eliciting safety-related aspects from existing norms and standards. 3. Facilitating effective analysis, tracing and reuse of change to prevent safety risks and nor- mative violations. The study identifies research gaps in advancements in NLP techniques for semantic and dis- course analysis, domain-specific word embedding models, human in the loop solutions and syntax- aware statement embedding that can benefit an automated retrieval of meaning for analysis of im- pact changes. Laying out the particular demands and constrains in SCS development, to answer the goals, this research addresses the identified constrains by exploring and outlining NLP methods and tools in a specific pipeline to improve aspects of the RE process for CIA. i