🤖 AI Summary
This work addresses the fragility of XML parsing and low editing success rates in natural-language-driven BPMN modeling. We propose an LLM-based process modeling approach leveraging a domain-specific JSON representation. Our core contributions are: (1) a lightweight, semantically explicit BPMN JSON Schema that replaces XML as the LLM’s input/output interface; (2) a quality evaluation framework combining graph edit distance (GED) and normalized GED to quantify structural fidelity, alongside a binary success metric to assess editing reliability. Experiments show that our method achieves process generation similarity comparable to XML-based baselines, while significantly improving editing success rate, inference speed, and robustness. The implementation is open-sourced, establishing a more efficient and resilient paradigm for LLM-powered business process modeling.
📝 Abstract
This paper presents BPMN Assistant, a tool that leverages Large Language Models (LLMs) for natural language-based creation and editing of BPMN diagrams. A specialized JSON-based representation is introduced as a structured alternative to the direct handling of XML to enhance the accuracy of process modifications. Process generation quality is evaluated using Graph Edit Distance (GED) and Relative Graph Edit Distance (RGED), while editing performance is evaluated with a binary success metric. Results show that JSON and XML achieve similar similarity scores in generation, but JSON offers greater reliability, faster processing, and significantly higher editing success rates. We discuss key trade-offs, limitations, and future improvements. The implementation is available at https://github.com/jtlicardo/bpmn-assistant.