UML Miner is a Visual Paradigm plug-in designed to support UML modeling education through AI-assisted formative feedback.
The tool records students’ modeling actions, generates a textual description of the current UML diagram, and allows learners to request feedback directly within the modeling environment. Feedback can be generated either by a standalone Large Language Model or by a Retrieval-Augmented Generation (RAG) configuration that grounds responses in instructional resources, assignment requirements, and example-based material.
UML Miner is intended to help novice learners reflect on their modeling decisions while constructing UML diagrams. Rather than acting as an automatic verifier that simply reports errors, the system provides contextualized suggestions, supporting evidence, and reflection questions that encourage students to compare their diagrams with the assignment requirements and improve their solutions iteratively.
The current version of UML Miner focuses on UML class diagram learning and supports empirical studies on AI-generated feedback, learner interaction, and modeling improvement. It can be used with the freely available Community Edition of Visual Paradigm for non-commercial educational and research purposes.
- How does it work?
For each Visual Paradigm project, UML Miner records the modeling history produced during diagram construction. The log includes modeling-related actions such as creating, deleting, or modifying UML elements, relationships, attributes, operations, multiplicities, and other diagram properties. Events related only to visual formatting or layout are excluded.
When a student requests feedback, UML Miner prepares three sources of contextual information:
the current UML diagram, converted into a structured textual description;
the modeling history recorded during diagram construction;
the student’s question, submitted either through predefined prompts or free-form text.
In the standalone LLM configuration, this information is sent directly to the language model. In the retrieval-augmented configuration, UML Miner first retrieves relevant instructional evidence from a knowledge base containing UML guidelines, assignment-related material, annotated examples, and representative feedback cases. The retrieved evidence is then combined with the learner context to generate feedback that is more closely aligned with the modeling task.