BPMN-Based Business Process Collaboration Modeling Using Large Language Models

This chapter explores the use of large language models (LLMs) to automate the creation of BPMN-based business process collaboration models from real-world Standard Operating Procedures (SOPs). The research addresses three key gaps in current LLM-based process modeling: the ability to represent processes defined within real-world SOPs, the generation of process collaboration models by defining message flows between process elements, and the inclusion of the data perspective in business process modeling. The proposed approach involves three main steps: process participant extraction, process entity extraction, and collaboration modeling. Each step leverages the role-playing capabilities of LLMs to identify participants, generate role-based process models, and define interactions using message flows. The approach is evaluated using several real-world SOPs made publicly available by the European Food Safety Authority (EFSA). The evaluation results indicate that the proposed approach is capable of creating nearly complete and correct data-annotated process collaboration models, but the modeling process must still have a human in the loop. The research concludes that while the approach captures the broader process and data perspectives of SOPs, it still yields inconsistent message flows, often due to defining send tasks without corresponding receive tasks and vice versa. This problem occurs because the send and receive tasks are defined for each participant in isolation, which can potentially be addressed by having another step within the modeling process that updates the FPD to ensure that every send task has a receive task and vice versa. The formalized process data, as well as the evaluation of this data, is made available on GitHub. This research demonstrates the viability of the proposed approach to defining the BPMN process model using an LLM-based decomposed modeling approach, but more testing is required to fully assess its capabilities.