Generative AI for Requirements Engineering (FFG Bridge Project 2024-2027)

Requirements Engineering (RE) is one of the most critical phases of a software project. Low-quality and incomplete requirements are often a major reason for software project failure. Due to a low level of automation support and complex information interchange and decision processes among stakeholders, RE is confronted with different open research issues. The major focus of the GenRE project is to analyze in which way large language models (LLMs) can help to significantly improve RE-related tasks such as the definition of requirements, quality assurance, reuse, and prioritization. The major idea of GenRE is to develop new techniques and methods that allow to integrate LLMs into RE-related processes to improve the automation and quality assurance support enabling a more efficent and sustainabile RE. In this context, the goals of GenRE are the following:

1) “LLM-based Requirements Elicitation & Specification“: develop new algorithms and techniques that support functionalities such as the automated (personalized) elicitation and generation of requirements, the automated derivation of requirements from unstructured sources, and the personalized support of stakeholders (e.g., by pro-actively stimulating discussions on the basis of explanations).

2) “LLM-based Analysis & Quality Assurance“: develop new algorithms and techniques that support the automated evaluation of requirements (with regard to criteria such as understandability and gender balance), automated analysis and explanation of hidden dependencies, and the recommendation of needed adaptations of requirements.

3) “LLM-enhanced Requirements Validation & Decision Making“: develop new approaches for the recommendation of group decision strategies and the impact-driven validation (e.g., business value and sustainability) and prioritization of requirements.

Project partners: TU Graz (Research Partner), Morgendigital (Company Partner), Innovation Service Network GmbH and Uniquare GmbH (external evaluation partners).

Publications:

• T.N.T. Tran, S. Polat-Erdeniz, A. Felfernig, S. Lubos, M. El Mansi, and V.M. Le, Less is More: Towards Sustainability-Aware Persuasive Explanations in Recommender Systems. In 18th ACM Conference on Recommender Systems (RecSys ’24), October 14–18, 2024, Bari, Italy. 2024.
• S. Lubos, A. Felfernig, T.N.T. Tran, D. Garber, M.E. Mansi, S.P. Erdeniz, and V.M. Le. Leveraging LLMs for the Quality Assurance of Software Requirements. In 32nd IEEE International Requirements Engineering 2024, RE@Next! Track, (RE@Next! 2024), Reykjavik, Iceland. 2024