Ontology-Based Layered Hybrid AI-Driven Knowledge Model for Personalized E-Learning

Synthesis: An ontology-based, layered hybrid knowledge model for personalized e-learning — a conceptual architecture paper by Tatyana Ivanova (2026) that classifies the knowledge required for personalization and proposes a layered knowledge-base architecture grounded in description logic. Its central move is replacing the classic ITS four-model architecture (domain, student, tutoring, interface) with systems of mapped ontologies, while adding procedural knowledge (rules), probabilistic/fuzzy knowledge (via fuzzy/probabilistic description-logic extensions), and implicit knowledge extracted through learning analytics and machine learning. Because ontologies alone are static and handle uncertainty poorly, the paper argues personalization requires combining them with teaching-strategy sequencing rules, analytics, and ML — and it contributes a metadata framework for describing, discovering, and reusing educational ontologies.

Key Findings

• Personalization needs more than ontologies. Ontologies are mostly static or slowly evolving and have limited handling of uncertainty, so effective personalized learning must combine them with teaching-strategy sequencing rules, learning analytics (to infer and predict learner needs), and machine learning for extracting implicit knowledge from educational data.
• Systems of mapped ontologies extend the ITS architecture. The paper differs from most prior work by proposing a system of mapped ontologies in place of each single ontology across the four classic ITS knowledge-base models (domain, student/tutoring, pedagogical, interface), together with storage for procedural knowledge (as rules) and imprecise/probabilistic knowledge (fuzzy or probabilistic description-logic extensions).
• ITS vs. IES distinction. An Intelligent Tutoring System is a specialized AI system focused on individual personalized tutoring; an Intelligent Educational System (IES) is a broader, multi-module ecosystem integrating tutoring, analytics, recommendation, and administrative decision-making. Ontologies in ITS enable cognitive tutoring; ontologies in IES mainly enable semantic integration.
• Ontology classification for reuse. The paper proposes a purpose-based classification of educational ontologies driven by the ITS model (tutoring-domain, pedagogy, or learner-profile), and a metadata framework with eight upper-level classes — ontologies, rules, mappings, data-driven (ML/analytics), ontology management tools, core descriptors, technical descriptors, and usage contexts — to support semantic search, evaluation, recommendation, and mapping, thereby reducing the time, effort, and cost of ontology development and evolution.

What this means for practice

• Designers. Match the representation to the knowledge type instead of defaulting to one: the architecture assigns crisp declarative knowledge to ontologies, procedural knowledge to rules, vague or uncertain knowledge to fuzzy or probabilistic ontologies, and implicit knowledge to analytics and machine learning.
• Designers. Prefer a system of small, mapped ontologies over one monolithic model spanning the four ITS knowledge bases (domain, student/tutoring, pedagogical, interface). Smaller ontologies are faster to develop, easier to validate, and more reusable across subjects, institutions, and levels.
• Developers. Reuse before you build: describe candidate ontologies, rules, mappings, and tools against the paper's eight-class metadata schema so discovery, evaluation, recommendation, and mapping can be automated rather than repeated on every project.
• Developers. Budget for mapping maintenance, not just authoring. Generate mapping candidates with matching tools (LogMap, AML, MELT), then validate with a human and check consistency with a description-logic reasoner (Pellet, HermiT), since equivalence and subsumption class mappings are the most frequent source of global inconsistency.
• Researchers. Validate a hybrid knowledge base on representative use cases and plan for expert review: even after automated ontology management, the authors' prior work shows expert evaluation is still required in each application context, and their dyslexia-tutoring and medical-tutoring cases show what a concrete instantiation requires.

Limitations

• Conceptual, not implemented: the paper contributes an architectural model and reports no working deployment, so it cannot support claims about runtime performance, development cost, or learning outcomes in real systems.
• Validation is limited to expert review by four researchers and two illustrative use cases (dyslexia tutoring, medical-student tutoring), largely carried over from the authors' earlier publications; no learner data or system evaluation is reported.
• The scope excludes several questions by design: the model offers no way to measure personalization benefits or interoperability and does not address standards variation across countries, institutions, and curricula, nor privacy and security in knowledge management.
• It assumes learners, pedagogical strategies, domain knowledge, and content can be explicitly formalized from observable interactions, which may not capture tacit, emotional, or context-dependent aspects of learning such as informal learning or complex instructor decision-making.

Citation

Ivanova, T. (2026). Ontology-based layered hybrid AI-driven knowledge model for personalized e-learning. Mathematics, 14(5), 808.