Reproduce “LLM-based Iterative Approach to Metamodeling in Automotive” 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
Nenad Petrovic, Fengjunjie Pan, Vahid Zolfaghari, Alois Knoll
Researchers at the Technical University of Munich introduce an iterative, LLM-based methodology to automate the construction and refinement of domain-specific metamodels for the automotive industry. This approach integrates textual requirements and human feedback to generate Ecore metamodels, reducing metamodel modification time from minutes to seconds compared to manual methods.
Key insights
- Large Language Models, particularly GPT-4o, can effectively abstract concepts and relationships from natural language to construct and iteratively refine formal, domain-specific metamodels.
- An iterative workflow combined with human feedback is crucial for building comprehensive and correct metamodels in complex domains, mitigating potential LLM inaccuracies ('hallucinations').
- Smaller, locally deployable LLMs show promise for capturing core structural elements (classes, compositions) but currently require additional model-to-model transformations and further refinement steps for complete Ecore metamodel generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.