A Model-Driven LLM-Assisted Approach for Generating Configuration Questionnaires from Business Process Family Models

Abstract

Variability models define configuration spaces in which stakeholders must make decisions before deriving concrete variants. Business process (BP) families constitute a representative application domain in which processes vary according to business requirements and contextual constraints. Modeldriven approaches support the specification of BP family models and the derivation of variants through model transformations; however, the increasing number and complexity of variability decisions make configuration difficult for end users. This paper explores how Large Language Models (LLMs) can support the generation of end-user questionnaires from BP family models. In our approach, questionnaires are treated as higher-level configuration models that capture user-oriented decisions and provide configuration information to automate variant derivation. The contribution is threefold: a metamodel-guided LLM generation pipeline for questionnaire-based configuration modeling; an exploratory empirical assessment of the generated models; and lessons on human-in-the-loop validation of AI-generated models. The results show the feasibility of combining modeldriven engineering and LLMs for configuration support, while also highlighting current limitations that must be addressed before deployment in industrial settings.