Hybrid validation of use cases: rule-based analysis, semantic NLP, and ontological grounding of LLM

Abstract

Use case validation plays an important role in software development. One of the main tasks of system analysts is to prevent errors in use cases before they lead to a dramatic increase in the cost of fixes. However, manual validation often suffers from incompleteness, ambiguity, logical contradictions, subjectivity of assessments, inconsistency of the process, and high time costs. According to a survey of system analysts (N=6), the average time for manual validation of a single use case is 50 minutes, and the proportion of defects caused by poor requirements reaches 38%. This paper presents a hybrid approach to automated use case validation that combines rule-based analysis (25 formal rules based on INCOSE and ISO/IEC 29148), semantic analysis using the Sentence-BERT model (similarity thresholds: 0.85 for duplicate detection, 0.40 for goal relevance), and ontological grounding of LLMs. A colored use case ontology (6 classes, 10 properties, 12 axioms) serves as a formal grounding mechanism for LLMs, ensuring traceability and determinism. The developed software module supports three input methods (structured form, JSON, file upload) and exports structured reports. Verification on 28 test cases confirmed 100% correctness, while validation on a real-world scenario demonstrated 100% defect detection. The implementation reduces validation time per use case from 50 minutes to 30 seconds and decreases requirements-related defects from 38% to 10%.