Entwicklung und Erweiterung eines Webservice-Prototyps zur automatisierten Bewertung von UML-Diagrammen unter Einsatz von Large Language Models

A webservice prototype was extended to automatically assess UML use‑case, class, and state diagrams by leveraging large language models. The system was evaluated with three models (GPT‑4.1‑mini, Claude Haiku 4.5, Qwen 3.5 35B) and three input formats (raw UXF data, parser‑derived representation, and image‑based diagrams), showing reliable classification for use‑case and class diagrams while state diagrams proved more difficult; the parser‑based format notably cut token usage without harming quality. Consequently, the approach is deemed suitable as a supportive grading aid rather than a fully autonomous evaluator.