Enhanced Structural Validation for UML Use Case Models with Configurable Rule Engine

Automated assessment of use case models in software engineering education requires both rigorous validation and pedagogical adaptability. This paper presents an AI-driven framework specifically designed for use case model evaluation, featuring a four-layer architecture and efficient batch processing via Sparx EA COM integration. The system combines spaCybased syntactic analysis—achieving $92 \%$ accuracy in detecting structural errors unique to use case models—with SBERT-based semantic validation tailored to use case requirements. Dynamic rule configuration is supported through JSON and Excel-based error categorization, enabling flexible adaptation to instructional needs. In a 30 -student pilot(Library Management System)use case models, the framework reduced grading time from 9 hours to 20 minutes and identified 3,050 violations, including 1482 missing pre/post conditions and $\mathbf{1 3 6}$ naming inconsistencies. Customizable lexical rules achieved $92.6 \%$ precision, though $30 \%$ of cases required instructor review due to linguistic ambiguity. Semantic validation provided actionable alerts within 0.4-0.85 confidence thresholds, despite some false positives. Overall, the framework delivers a $27 \times$ speedup and high accuracy in use case model assessment, significantly enhancing UML education. Code and case data are released under CC BY-NC 4.0 for non-commercial research.