Intelligent systems draw much of their reliability from the quality of their ontologies; however, manual ontology assessment remains patchy, time-consuming, and difficult to scale. To address these limitations, this paper proposes a domain-independent, machine-learning-driven framework for ontology quality assessment and improvement in the Semantic Web. The framework combines structural, semantic, and documentation metrics with supervised learning models to predict quality issues and recommend targeted refinements through a four-phase workflow comprising ML model development, metric definition, automated improvement, and empirical evaluation. The approach is validated on educational knowledge graphs using 1500 ontology modules from the EDUKG repository, including a 100-module expert-annotated gold set (κ = 0.82). Experimental results show structural precision of 93.5% and semantic precision of 90.2%, with overall F1-scores close to 90%, while reducing ontology development time by 42% and quality assessment time by 65%. These findings demonstrate that coupling ML with structured quality metrics substantially enhances ontology reliability while preserving pedagogical and operational relevance in educational settings. Although empirical validation is conducted in the education domain, the modular and ontology-agnostic architecture can be adapted to other knowledge-intensive domains through retraining and domain-specific calibration, offering a reproducible foundation for continuous ontology quality improvement in Semantic Web applications.
Jaziri et al. (Sat,) studied this question.