V. Muttillo, R. Eramo, R. Rubei and L. Berardinelli: Coupling LLMs and Model-Driven Engineering to Support Synthetic Generation of BPMN Artifacts, ACM/IEEE 28th International Conference on Model Driven Engineering Languages and Systems Companion (MODELS-C), pp. 556-565, Grand Rapids, MI, USA, October 5-10, 2025. Doi: 10.1109/MODELS-C68889.2025.00079
The rise of large language models (LLMs) has led to new opportunities and challenges, particularly in model-driven engineering (MDE), where they are utilized to automate various modeling tasks. However, current research highlights some limitations that hamper the LLMs’ effectiveness, e.g., producing formatting errors in the generated modeling artifacts or semantic hallucination in domain modeling. Modeling business process management (BPM) applications presents significant challenges to end-users, ranging from adhering to OMG standards to ensuring data privacy. Although some approaches have started investigating the intersection between LLMs and BPM modeling, a completely automated process is still far from being realized. The main blocking issue is represented by the scarcity of modeling data, which is needed to train both LLMs and traditional modeling assistants. Being built on our previous work, we adopt an MDE framework called BP-MASTER-LLM, which is devoted to supporting the specification of BPM models by combining a set of automated modeling tools, i.e., model-event recorder and modeling assistant. Our findings reveal that the prominent LLMs can generate synthetic content even though human models and operations represent a better source of training data for the tested modeling assistant; nonetheless, if instructed properly, LLMs can represent a solution to overcome the data scarcity issue.