Scholarlink

This study investigates the zero-shot capabilities of Large Language Models (LLMs) in generating Business Process Model and Notation (BPMN) diagrams for Robotic Process Automation (RPA). The research evaluates the performance of LLMs, specifically GPT-4o, in creating valid BPMN XML diagrams compatible with Camunda, a widely used process automation platform. A comprehensive dataset comprising 108 diagrams across four categories of workflow complexity was generated and expert-labeled. The study assesses the quality of the generated diagrams based on six criteria: validity, completeness, detail, arrow correctness, readability, and proper task delegation. The results provide insights into the strengths and limitations of LLMs in automating BPMN diagram creation, highlighting areas where these models excel and identifying potential focuses for future fine-tuning efforts. The findings suggest that while LLMs show promise in generating valid and readable BPMN diagrams for basic and intermediate workflows, challenges remain in representing highly complex processes, indicating areas for future research and development in AI-driven process modeling.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automating Automation: Using LLMs to Generate BPMN Workflows for Robotic Process Automation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Carlos Toxtli,
  • Wangfan Li

摘要

This study investigates the zero-shot capabilities of Large Language Models (LLMs) in generating Business Process Model and Notation (BPMN) diagrams for Robotic Process Automation (RPA). The research evaluates the performance of LLMs, specifically GPT-4o, in creating valid BPMN XML diagrams compatible with Camunda, a widely used process automation platform. A comprehensive dataset comprising 108 diagrams across four categories of workflow complexity was generated and expert-labeled. The study assesses the quality of the generated diagrams based on six criteria: validity, completeness, detail, arrow correctness, readability, and proper task delegation. The results provide insights into the strengths and limitations of LLMs in automating BPMN diagram creation, highlighting areas where these models excel and identifying potential focuses for future fine-tuning efforts. The findings suggest that while LLMs show promise in generating valid and readable BPMN diagrams for basic and intermediate workflows, challenges remain in representing highly complex processes, indicating areas for future research and development in AI-driven process modeling.