Reproduce “Leveraging Large Language Models for Enhanced Process Model Comprehension” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Humam Kourani, Alessandro Berti, Jasmin Hennrich, Wolfgang Kratsch, Robin Weidlich, Chiao-Yun Li, Ahmad Arslan, Daniel Schuster and 1 more
This research introduces a framework leveraging Large Language Models to enhance the interpretability of complex Business Process Model and Notation (BPMN) diagrams, demonstrating that JSON and Simplified XML (SXML) are effective abstraction formats. The study shows specific prompting strategies improve LLM comprehension, with GPT-4o performing well on domain-specific questions and open-source models exhibiting comparable capabilities.
Key insights
- JSON and Simplified XML (SXML) abstraction formats offer the best trade-off between output quality and token usage for LLM comprehension of BPMN models.
- Non-Technical Abstraction, Few-Shot Learning, and Negative Prompting significantly enhance LLM comprehension, while Role Prompting and Chain of Thoughts have minimal impact.
- State-of-the-art LLMs, including open-source options like WizardLM-2-8x22B and Mixtral-8x22B-Instruct, demonstrate comparable performance to commercial solutions like GPT-4o for BPMN interpretation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.