AUG: an Interactive Tool for Clarifying and Generating UML Models Based on Large Language Models
Jiaming Lu, P. Sun, Yi‐Ping Phoebe Chen, Gege Yin, Peng Ye · 2025
Requirements modeling is crucial in software development, yet manual UML modeling is time-consuming and error-prone. Existing Large Language Model (LLM)based approaches often rely on indirect methods like PlantUML parsing and lack adaptability. To address these challenges, we propose AUG (Automated UML Model Generation), an interactive tool powered by the GLM4UML model for automating UML modeling, particularly for class diagrams, use case diagrams, and sequence diagrams. AUG integrates dynamic editing, quality assessment, and feedback mechanisms, forming an automated evaluation and optimization paradigm. We constructed a dataset of 10,000 entries and fine-tuned GLM4-9B using prompt-based techniques to enhance model performance, creating the GLM4UML model. Experimental results on a real-world test set show that AUG achieves precision, recall, and F1 scores of$78.68 \%, 67.37 \%$, and 72.59 %, respectively, outperforming four open-source LLMs. While AUG's accuracy is slightly lower than the proprietary ChatGPT-4o, it demonstrates superior recall, F1 score, and overall stability. A questionnaire study further highlights AUG's usability and efficiency, with high user satisfaction reported. The repository, accessible at https://github.com/XIAOLingQ/AUG, contains the code, datasets, and evaluation data. The pretraining details are available at https://huggingface.co/XIAOLQ/GLM4UML and a demo video can be found at https://youtu.be/kHbCPK6kOag.