AI Augmented Requirements Engineering in the AIDOaRT Project: NLP Techniques and Language Models to Encode Requirements Text Semantically for the Railway Industry
Alessandra Bagnato, Bilal Said · ACM SIGAda Ada Letters · 2025
Requirements engineering (RE) remains a complex and crucial preliminary phase in the life cycle of a Cyber- Physical System (CPS) design and development process. In CPS, such as railway and automotive systems, the requirements of a single system, often expressed and handled in unstructured text format, may exceed several thousands of clauses, with technical jargon, multitude of references to external safety and functional standards, extensive cross-references and dependencies. This makes RE activities, such as requirements analysis, validation and allocation, time consuming and error prone for system engineers. This paper contributes with Natural Language Processing (NLP) Techniques and Language Models (LM) to encode requirements text semantically. It describes NLP4RE, a tool by Softeam for requirements semantic similarity search. It has been applied for a use case from the railway industry within the AIDOaRT project. In this use case, system engineers get valuable insights on new bid documents by retrieving semantically similar clauses from previous projects.