About

Arnon Sturm is a Professor in the Faculty of Computer and Information Science at Ben-Gurion University of the Negev, Israel, where he has been a faculty member since 2005. He holds a B.Sc., M.Sc., and Ph.D. from the Technion — Israel Institute of Technology. Prior to his academic career, he gained industry experience at the Aerospace and Telecommunication companies and served as a researcher at IBM Haifa Research Lab. His research interests span requirements engineering, conceptual modeling, and software development methodologies, with a particular focus on the derivation of domain models from user requirements, the empirical evaluation of requirements notations and techniques, and the application of machine learning and artificial intelligence to software engineering tasks. In recent years, he also applied NLP techniques to law domain. He serves on the editorial board of Business & Information Systems Engineering and has served as a guest editor for Software and Systems Modeling. He is an active member of the requirements engineering and conceptual modeling communities, regularly serving as co-chair and program committee member at venues including CAiSE, ER, and ICSE. He chaired both the Information Systems Engineering and the Software Engineering Programs and was involved in many entrepreneurship initiatives at Ben-Gurion University of the Negev.

Abstract

From Requirements to Domain Models: A Journey Through Human and Machine Derivation

Domain models play a crucial role in software development, as they provide means for communication among stakeholders, for eliciting requirements, and for representing the information structure behind a system or at the basis of model-driven development. Yet, the question of how to systematically derive domain models from existing requirements remains open. To address that question, we report on a series of studies that examined (1) the impact of notations for requirement specifications, specifically user stories versus use cases, by means of controlled and quasi-experiments involving both undergraduate and graduate students; our findings suggest that while user stories tend to yield more complete and valid models in time-constrained settings, domain complexity and the derivation process prove to be stronger determinants of quality in more realistic settings; (2) the way people derive domain models, by designing and evaluating example-based guidelines that improve model completeness and validity, particularly for complex domains; (3) the effectiveness of automatic approaches for deriving domain models, including rule-based, machine-learning, and LLM-based approaches, which we benchmarked against human achievable performance across nine collections of user stories, our finding indicate that no automated approach consistently matches human performance, although a tuned machine learning classifier comes closest; and finally (4) the implications of these findings on the adoption of LLM-based approaches by human analysts. We elaborate on each study and conclude with open challenges and directions for future research.

About Current Trends in Computer Science

This lecture is part of the Current Trends in Computer Science Lecture Series by the TU Wien Informatics Doctoral School, where renowned Guest Professors hold public lectures every semester. If you are studying with us, the lecture series can be credited as an elective course for students of master’s programs of computer science: 195.072 Current Trends in Computer Science. Additionally, you can join courses held by this year’s Guest Professors of our doctoral colleges and the TU Wien Informatics Doctoral School.

Curious about our other news? Subscribe to our news feed, calendar, or newsletter, or follow us on social media.