Prompt 2 — Ontology-Grounded Reasoning Systems
Conduct a comprehensive review of ontology-based artificial intelligence, semantic web technologies, knowledge representation, description logics, OWL/RDF frameworks, enterprise ontologies, and model-driven engineering. Evaluate how formal ontologies can be integrated with large language models, retrieval systems, and agentic architectures to create reasoning systems capable of architectural inference, impact analysis, and mission-level decision support.
Generated on: June 07, 2026
Comprehensive Literature Review: Ontology-Grounded Reasoning Systems
1. Foundations of Ontology-Based Artificial Intelligence
1.1 Historical Context and Core Principles
Ontologies have emerged as fundamental components of knowledge systems, providing explicit, formal specifications of shared conceptualizations within domains of interest. The classical definition establishes ontologies as abstract, simplified views of worlds including objects, concepts, and their interrelationships [1]. Since the 1970s, ontologies have played increasingly vital roles in knowledge representation, enabling unambiguous structuring of domains, facilitating knowledge sharing and reuse, and supporting automated reasoning [1]. The transition from manual knowledge engineering to more scalable, automated approaches has accelerated with advances in semantic web technologies and integration with machine learning paradigms.
The fundamental role of ontologies in knowledge representation extends beyond academic interest to practical enterprise applications. As systems become increasingly complex and interconnected, the ability to formally represent and reason about domain knowledge has become critical for organizational success [2]. Ontologies serve as conceptual frameworks that bridge raw data and machine cognition, supporting semantic enrichment of AI systems [2]. This capability proves particularly valuable in healthcare, natural language processing, and semantic web applications, where semantic understanding directly impacts system reliability and decision quality.
1.2 Description Logics as Formal Foundations
Description Logics (DLs) constitute a fundamental family of languages for formal knowledge representation, providing decidable fragments of first-order logic that balance expressive power with computational tractability [3]. DLs enable efficient reasoning tasks including satisfiability checking and entailment decision, forming the logical underpinning of the W3C Web Ontology Language (OWL) standard [1]. The mathematical foundations of DLs ensure that systems built upon these languages possess well-defined semantics, enabling sound and complete reasoning.
The relationship between description logics and ontology development is symbiotic. DLs facilitate various reasoning tasks through their formal structure, yet the challenge of explaining reasoning results to humans requires integration with cognitive science principles [3]. Graph-based representations of DL models, where nodes and edges represent concepts and relationships, provide visual foundations for understanding complex semantic structures. This formal grounding becomes particularly important when ontologies are deployed in mission-critical systems requiring explainability and auditability.
1.3 Ontology Engineering Methodologies
Systematic approaches to ontology construction have evolved considerably, with established methodologies providing structured frameworks for knowledge engineers. The Uschold-Grüninger methodology and related approaches establish formal processes for identifying domain concepts, defining relationships, and ensuring consistency [4]. Modern ontology engineering integrates these traditional approaches with contemporary tools like Protégé, WebProtégé, and VocBench, enabling collaborative development and semantic validation. For educational and organizational contexts, hybrid approaches combining Guarino&#39;s formal classifications with Uschold-Grüninger&#39;s pragmatic methods offer scalable, context-aware ecosystems [4].
Quality assessment of ontologies requires multidimensional evaluation frameworks encompassing logical consistency, semantic expressiveness, and practical utility [5]. Tools like HermiT reasoners provide automated consistency checking, while SPARQL Protocol and RDF Query Language (SPARQL) competency questions validate ontologies&#39; ability to represent required knowledge [6]. The integration of metaheuristic optimization techniques with ontology development enhances efficiency and supports adaptive learning processes within smart information systems, though maintaining balance between model complexity and computational efficiency remains an ongoing challenge.
1.4 Semantic Interoperability and Data Integration
Semantic interoperability—the ability to exchange meaningful information across diverse systems—constitutes a central driver for ontology adoption in enterprise environments [7]. Ontologies enable systems to interpret data consistently despite heterogeneous underlying schemas and formats, particularly crucial when integrating legacy systems with modern cloud-native architectures [8]. The use of standardized ontology vocabularies from frameworks like CIDOC CRM and SKOS ensures that semantic representations remain interpretable across organizational and disciplinary boundaries [6].
The challenges of semantic data understanding in hybrid infrastructures—combining structured, semi-structured, and unstructured sources—drive innovation in layered knowledge graph architectures [9]. These frameworks integrate deterministic schema metadata with flexible semantic enrichment derived from documentation and business glossaries, creating unified semantic layers that bridge business meaning and technical metadata [9]. Success in semantic integration requires addressing data quality, consistency, and maintenance challenges, with automated approaches increasingly complementing manual curation to achieve scalable knowledge management.
2. Semantic Web Technologies and Knowledge Representation Frameworks
2.1 RDF/RDFS: Graph-Based Data Models
The Resource Description Framework (RDF) establishes a foundational data model for semantic web technologies, representing information as edge- and vertex-labeled directed graphs where triples (subject, predicate, object) constitute atomic units of meaning [10]. RDF&#39;s graph-based structure naturally accommodates complex relationships and distributed data sources, providing a flexible foundation for knowledge representation across organizational boundaries. By representing data as interconnected nodes rather than rigid relational schemas, RDF enables more intuitive representation of domain semantics and supports exploratory reasoning over diverse knowledge types.
RDF Schema (RDFS) extends RDF with vocabulary and semantic mechanisms for defining classes, properties, and their relationships, providing an intermediate layer between RDF&#39;s simplicity and OWL&#39;s expressiveness [10]. This layered approach enables organizations to adopt semantic technologies incrementally, starting with basic RDF graphs and progressing toward richer ontological representations as organizational capabilities mature. The standards-based nature of RDF and RDFS—sanctioned by W3C—ensures interoperability across platforms and tools, facilitating knowledge sharing across enterprise boundaries.
2.2 OWL Ontology Language and Expressiveness
The Web Ontology Language (OWL), based on description logics, provides sophisticated mechanisms for expressing complex ontological constraints while maintaining computational tractability [11]. OWL&#39;s multiple profiles—including OWL 2 DL with full expressiveness and OWL 2 QL optimized for data access—enable organizations to select appropriate expressiveness levels for their specific requirements [1]. By supporting formal constraints, inheritance hierarchies, and reasoning axioms, OWL enables ontologies to capture subtle domain semantics that simpler representations cannot express.
The potential of OWL-based knowledge graphs for neuro-symbolic systems has gained substantial recognition, particularly as researchers recognize the benefits of formal logical reasoning capabilities for constraining and validating neural models [12]. OWL&#39;s ability to express complex logical rules and constraints enables hybrid systems that combine neural learning with symbolic reasoning, supporting applications requiring both adaptability and interpretability [12]. The integration of OWL with emerging AI technologies demonstrates how formal semantic standards can enhance rather than constrain modern intelligent systems.
2.3 SPARQL Query Language and Inference
SPARQL Protocol and RDF Query Language (SPARQL) enables powerful querying capabilities over RDF graphs, supporting complex multi-hop reasoning, filtering, and aggregation operations [10]. Unlike traditional SQL queries operating over relational schemas, SPARQL queries traverse semantic relationships encoded in RDF graphs, enabling reasoning tasks that capture domain logic implicitly expressed through ontological structures. This query capability proves particularly valuable for knowledge discovery, impact analysis, and decision support applications requiring multi-dimensional reasoning.
Ontology-mediated query answering (OMQA) frameworks leverage SPARQL to provide virtual access to relational databases through ontological mappings, enabling organizations to query legacy systems using semantic concepts [13]. This approach avoids the need to materialize entire databases as RDF, instead translating SPARQL queries into optimized relational queries through query rewriting techniques. The resulting systems maintain data currency while providing semantic access layers that insulate users from underlying technical complexity.
2.4 Linked Data Principles and Knowledge Graphs
Linked data principles establish practices for publishing and connecting structured data across the web, enabling the creation of globally interconnected knowledge graphs [10]. By using URIs as universal identifiers, explicit ontological relationships, and standard formats like RDF and OWL, linked data creates networks of meaning that span organizational and disciplinary boundaries. The resulting knowledge graphs support advanced forms of reasoning, entity disambiguation, and cross-domain inference not possible within isolated data silos.
The relationship between ontologies and knowledge graphs represents one of the most consequential developments in semantic technologies. Knowledge graphs—large-scale networks of entities and their relationships enriched with semantic metadata and ontological constraints—have become central infrastructure in enterprises, supporting applications from search and recommendation to knowledge discovery and decision support [10]. The formal semantics provided by ontologies ensures that knowledge graphs maintain consistency and enable sound inference, distinguishing them from purely statistical embeddings or unstructured networks.
3. Enterprise Ontologies and Model-Driven Engineering
3.1 Enterprise Architecture Frameworks
Enterprise architecture (EA) frameworks provide structured approaches for designing and evolving complex organizational systems, increasingly integrating semantic technologies to enhance knowledge management and decision support. The alignment of EA practices with semantic web technologies creates opportunities for more dynamic, responsive systems capable of capturing both structural and behavioral aspects of enterprises [14]. By treating enterprise architecture as a dynamic knowledge system incorporating multiple layers of organizational knowledge, organizations can achieve adaptive governance mechanisms.
The integration of ontological knowledge modeling into academic and organizational processes has demonstrated substantial benefits for consistency, interoperability, and governance [4]. Case studies from multi-school curriculum governance illustrate how semantic web-driven EA frameworks reduce documentation inconsistencies by up to 24%, while query efficiency improves by approximately 50% through structured semantic representations [14]. These improvements translate directly into operational benefits: reduced audit overhead, faster decision-making, and improved alignment between organizational strategy and technical implementation.
3.2 Model-Driven Engineering (MDE) Approaches
Model-driven engineering represents a paradigm where abstract models serve as primary artifacts, with executable systems and documentation generated through automated transformations. MDE approaches have demonstrated effectiveness across diverse domains, with meta-modeling and model transformation emerging as core techniques for managing system complexity. The combination of MDE with ontological semantics creates particularly powerful frameworks, where formal domain models capture both structural constraints and semantic relationships.
The synergy between model-driven engineering and semantic web technologies has enabled novel approaches to system development, particularly in contexts requiring both formal verification and semantic transparency. Recent work demonstrates how transformations can leverage structured metadata to support consistency checking and systematic development processes. This approach combines the expressiveness of formal specifications with practical implementation support, reducing development effort while maintaining consistency and traceability.
3.3 Domain-Specific Languages and Metamodeling
Domain-specific languages (DSLs) provide tailored notations for expressing concepts and constraints within particular domains, complementing general-purpose modeling languages with domain-appropriate abstractions. DSL development has demonstrated maturity in the systems modeling field, with established practices for creating languages that balance expressiveness with usability. This tool ecosystem demonstrates effective approaches to enabling domain experts to drive system development without requiring deep technical expertise.
Specialized approaches in domain-specific contexts exemplify how formal semantics can enable automated reasoning and knowledge management. These frameworks enable domain experts and data scientists to define complex operations using shared ontologies, with systematic transformations maintaining consistency and correctness. When combined with appropriate semantic foundations, domain-specific approaches enable organizations to maintain intellectual capital in machine-processable form while democratizing access to sophisticated modeling capabilities.
3.4 Knowledge Management in Enterprise Systems
Knowledge management systems increasingly incorporate ontological frameworks to organize, represent, and make accessible the intellectual capital essential for organizational success. Ontology-based approaches to knowledge management support not only documentation and retrieval but also sophisticated reasoning capabilities including impact analysis, consistency checking, and decision support. By formalizing domain knowledge through ontologies, organizations transform tacit expertise into explicit, machine-processable representations that can be applied consistently across operational and strategic processes.
The integration of semantic technologies with enterprise systems addresses persistent challenges in knowledge accessibility, consistency, and reuse. AI-driven enterprise architectures combining ontological knowledge representation with metadata management demonstrate improvements in interoperability efficiency, metadata accuracy, and anomaly detection compared to traditional integration approaches. These quantified improvements highlight the practical value of investing in ontological infrastructure as part of enterprise digital transformation initiatives.
4. Integration of Ontologies with Large Language Models and Retrieval Systems
4.1 Ontology-Grounded LLM Architectures
The integration of formal ontologies with large language models represents a significant advance in AI systems, addressing fundamental limitations of LLMs while leveraging their considerable strengths in natural language understanding and generation [15]. Ontology-grounded frameworks enforce structured semantic grounding during reasoning processes, mitigating hallucinations and improving factual accuracy. The dramatic results achieved through ontology-grounded clinical question answering—achieving 98% accuracy compared to 37% for baseline ChatGPT-4 and reducing hallucination rates from 63% to 1.7%—demonstrate the transformative potential of this integration [15].
Hybrid approaches that combine LLMs with ontological reasoning create systems capable of both learning from data and reasoning with formal constraints. By integrating OWL ontologies with symbolic reasoners like HermiT, systems can detect inconsistencies between LLM outputs and domain knowledge, then generate corrective feedback iteratively. This closed-loop approach maintains the fluency and contextual awareness of LLMs while imposing the rigor and consistency requirements of formal systems.
4.2 GraphRAG and Retrieval-Augmented Generation
Retrieval-Augmented Generation (RAG) has emerged as a powerful approach for customizing LLMs with domain-specific knowledge, seamlessly integrating external knowledge bases to enable real-time access to specialized expertise. Graph-based retrieval-augmented generation (GraphRAG) extends traditional RAG by leveraging structured knowledge representations, explicitly capturing entity relationships and domain hierarchies to enable context-preserving knowledge retrieval with multi-hop reasoning capabilities. This advancement addresses critical limitations of flat text retrieval, enabling systems to answer complex queries requiring contextual reasoning across distributed knowledge sources.
The practical deployment of GraphRAG demonstrates substantial improvements in knowledge retrieval and reasoning quality. Unified frameworks integrating knowledge bases with graph representations and semantic search improve answer relevance compared to standalone LLM-based retrieval pipelines. By combining graph construction, hybrid reasoning, and interactive visualization, such frameworks provide scalable, explainable foundations for intelligent knowledge assistants supporting decision-making across enterprise environments.
4.3 Knowledge Graph Construction with LLMs
Large language models have demonstrated remarkable capabilities for extracting structured knowledge from unstructured text, automating construction of domain-specific knowledge graphs that would otherwise require extensive manual curation [16]. The integration of LLM-powered entity extraction, relationship inference, and semantic enrichment enables rapid development of knowledge graphs capturing complex domain knowledge. When combined with ontology-aligned RDF/OWL schema generation and multi-LLM consensus validation, these systems produce interpretable, SPARQL-compatible knowledge graphs grounded in domain-specific reasoning [16].
The application of multi-agent LLM systems to knowledge graph construction has expanded possibilities for handling heterogeneous data sources and complex knowledge integration [17]. Frameworks employing adaptive ontology mapping using APIs for major medical standards (SNOMED CT, ICD-11) ensure semantic interoperability while processing data from comprehensive surveys [17]. The integration of LLMs with ontology tools like Protégé demonstrates how automation and human expertise can combine to produce high-quality, clinically relevant knowledge graphs at scale.
4.4 Hallucination Mitigation Through Semantic Grounding
LLM hallucinations—the generation of plausible-sounding but factually incorrect information—constitute a major impediment to enterprise deployment in domains requiring accuracy and accountability. Ontology-grounded approaches address this challenge by enforcing factual consistency against formal domain knowledge [15]. When LLM outputs are constrained to conform to domain ontologies, the systems achieve both high accuracy and interpretability, as incorrect responses violate explicit semantic constraints rather than being masked by statistical uncertainty.
The mechanisms underlying hallucination mitigation through semantic grounding involve multiple reinforcing factors. Formal ontologies encode ground truth about domain concepts and relationships, enabling systems to validate LLM outputs against authoritative knowledge. Semantic web technologies like SPARQL enable efficient querying of these constraints, while symbolic reasoners verify logical consistency. The combination creates multi-layered validation systems where LLMs generate candidate responses that are then filtered through semantic constraints, ensuring that only factually grounded outputs are presented to users.
5. Neuro-Symbolic Reasoning Systems and Hybrid Intelligence
5.1 Neural-Symbolic Integration Frameworks
Neuro-symbolic artificial intelligence addresses the fundamental divide between connectionist approaches excelling at pattern recognition and symbolic systems excelling at logical reasoning. By integrating neural networks with formal logic, neuro-symbolic systems achieve capabilities unavailable to either paradigm alone: the robust learning abilities of neural networks combined with the transparency and reasoning power of symbolic systems. The promise of neuro-symbolic integration extends to addressing the knowledge acquisition bottleneck, enabling systems to learn high-quality representations suitable for formal reasoning.
The successful integration of symbolic and subsymbolic paradigms requires careful architectural design to enable seamless knowledge exchange and reasoning. Approaches grounding first-order logic through differentiable operations enable neural networks to be trained with formal logical constraints, ensuring that learned representations remain interpretable and maintain consistency with domain knowledge. Experimental evaluations demonstrate that logic-grounded approaches achieve state-of-the-art performance while maintaining logical consistency on benchmark tasks.
5.2 Logic Tensor Networks and Differentiable Reasoning
Logic-grounded frameworks provide mathematical foundations for seamlessly integrating neural learning with symbolic reasoning by grounding logical formalisms in differentiable operations. By relaxing logical operators into continuous functions and mapping logical predicates to learned embeddings, such approaches enable end-to-end training of hybrid systems maintaining both learning capability and logical structure. This approach has enabled novel applications including improved zero-shot learning through incorporation of class hierarchies and robust inductive biases.
The compilation of description logic ontologies into probabilistic circuits provides another avenue for integrating neural and symbolic computation at scale [18]. By encoding DL ontologies as differentiable circuits, systems achieve three distinct capabilities: generation of synthetic datasets capturing ontology semantics, efficient GPU-accelerated deductive reasoning with runtimes substantially faster than traditional reasoners, and neuro-symbolic classification producing predictions consistent with ontological constraints [18]. These advances demonstrate that the marriage of logic and learning can yield practical systems addressing real-world challenges in data-constrained and knowledge-rich domains.
5.3 Knowledge Representation in Hybrid Architectures
Hybrid architectures combining neural perception with symbolic reasoning face fundamental questions about knowledge representation and the interfaces enabling effective communication between paradigms. Adaptive frameworks with dynamic contextual reasoning address this challenge through contextual embedding layers, hierarchical knowledge graphs encoding multi-level relationships, and adaptive reasoning engines performing context-aware logical inference. This architecture enables systems to move beyond pattern recognition toward deeper semantic understanding and dynamic adaptability.
The representation of knowledge across neural-symbolic systems must accommodate both the statistical learning strengths of neural components and the formal reasoning requirements of symbolic components. Applications in specialized domains requiring integration of diverse data types demonstrate that effective neuro-symbolic systems employ layered representations where neural networks extract feature representations from raw data, symbolic layers encode domain knowledge and reasoning rules, and integration mechanisms enable bidirectional feedback. This approach has achieved notable adoption, with systems generating both predictions and transparent rule-based explanations enhancing accuracy and interpretability.
5.4 Explainability and Interpretability Mechanisms
Explainability constitutes a central motivation for neuro-symbolic approaches, as formal logic enables systems to justify decisions through traceable reasoning chains rather than opaque numerical weights. By augmenting large language models with symbolic reasoning layers, hybrid systems generate natural language explanations accompanied by logical justification paths, providing transparency vital for regulated domains and high-stakes applications. Experimental evaluations demonstrate substantial improvements in logical coherence, reasoning accuracy, and explanation fidelity compared to purely neural baselines.
The integration of symbolic knowledge graphs with neural perception creates opportunities for generating human-interpretable explanations grounded in formal domain semantics. In autonomous systems, neuro-symbolic approaches enable accurate interpretation of complex scenes while providing clear explanations for decisions. In clinical applications, the combination of symbolic reasoning with neural perception generates diagnoses accompanied by readable rule-based explanations, significantly enhancing clinician trust and enabling verification of system reasoning against established medical knowledge. These advances position neuro-symbolic systems as essential for AI deployment in domains where explainability directly impacts system acceptance and regulatory compliance.
6. Agentic Architectures and Mission-Level Decision Support
6.1 Multi-Agent Systems with Ontological Grounding
Multi-agent systems represent an architectural paradigm for organizing complex distributed problem-solving, with ontologies providing shared semantic foundations enabling effective inter-agent communication and coordination. The integration of large-scale ontological knowledge graphs with multi-agent systems has enabled novel forms of autonomous scientific discovery, where agents reason over interconnected domain concepts to identify previously unrecognized relationships. Frameworks demonstrating multi-agent reasoning over domain ontologies reveal hidden interdisciplinary connections, achieving discovery capabilities exceeding traditional research methods.
Semantic grounding through shared ontologies enables agents to operate with common understanding of domain concepts, eliminating ambiguities that would otherwise require extensive communication overhead. The systematic taxonomy of agentic reasoning frameworks demonstrates the breadth of architectural possibilities across single-agent methods, tool-based methods, and multi-agent approaches. Applications across scientific discovery, healthcare, software engineering, and other domains validate that ontologically grounded multi-agent systems can address complex problems requiring coordination, knowledge synthesis, and adaptive learning.
6.2 Architectural Inference and Impact Analysis
Reasoning about architectural systems—predicting how changes propagate through complex interdependencies—requires formal semantic representations capturing structural relationships and functional dependencies. The integration of dynamic multi-layer knowledge graphs with intelligent decision-making methods enables organizations to move from reactive impact analysis to proactive system management. By constructing hierarchical knowledge architectures capturing multiple representation levels, systems can automatically parse change propagation paths, identify critical elements related to engineering changes, and combine expert experience with data-driven pattern identification.
Graph-aware learning approaches demonstrate how machine learning over graph-based representations enables prediction of system impacts with high accuracy. By leveraging graph neural networks trained on enterprise data, systems identify complex interaction patterns that humans cannot reliably predict manually. The superior performance of graph-aware models compared to traditional approaches demonstrates that ontologically structured representations capture essential domain semantics enabling more effective reasoning about system dynamics.
6.3 Enterprise Decision-Making Systems
Transforming enterprise decision-making through AI requires integration of knowledge graphs with formal reasoning, enabling systems to identify meaningful relationships among variables and support evidence-based decision-making. The integration of knowledge representation with structured reasoning establishes transparent decision frameworks ensuring that predictions possess both accuracy and justifiability. Evaluations on enterprise datasets demonstrate that structured reasoning approaches outperform mainstream deep learning approaches in error control and generalization performance, showing higher robustness across different conditions.
The construction of intelligent decision-making systems integrating knowledge graphs with domain-specific reasoning has demonstrated substantial improvements in organizational performance. When evaluated on complex scenarios, sophisticated reasoning achieves high accuracy, maintains low latency for real-time applications, and generates interpretable policy recommendations. These performance metrics enable decision support for complex operational scenarios where decision-makers require immediate access to interpretable reasoning about system states and meaningful relationships.
6.4 Real-World Applications and Case Studies
The application of ontology-grounded reasoning systems to real-world enterprise challenges has generated compelling evidence for their transformative potential. In financial services, ontological knowledge graphs enable semantic automation of regulatory compliance reporting, with sophisticated frameworks demonstrating advantages in real-time reporting, semantic interoperability, and auditability [8]. Such frameworks employ ontology-based data ingestion, knowledge graph building, rule-based inference engines, and real-time dashboards to support complex financial reasoning. Implementation outcomes show accuracy improvements, reduced latency, and regulatory transparency, with significant reductions in data reconciliation errors compared to traditional systems [8].
The integration of LLMs with knowledge graphs has enabled novel approaches to complex domain reasoning across healthcare and other specialized fields. Applications in medical diagnosis, materials discovery, and equipment maintenance demonstrate that ontology-grounded reasoning systems enable autonomous systems to reason about complex domains while maintaining full transparency and consistency with formal domain knowledge.
Conclusion
Ontology-grounded reasoning systems represent a fundamental shift in how artificial intelligence can be deployed to support human decision-making and automate complex reasoning tasks. By integrating formal semantic representations through ontologies with advanced learning and reasoning techniques—from large language models to neuro-symbolic systems to multi-agent architectures—organizations can create AI systems that combine the strengths of symbolic logic, statistical learning, and neural pattern recognition. The evidence presented throughout this review demonstrates that ontological grounding addresses persistent AI limitations including hallucinations, interpretability, knowledge integration, and consistency.
The synergy between semantic web technologies (RDF, OWL, SPARQL), formal knowledge representation (description logics, ontology engineering), and emerging AI paradigms (LLMs, neuro-symbolic systems, agentic architectures) creates unprecedented opportunities for building trustworthy, explainable, and effective reasoning systems. Enterprise adoption increasingly demonstrates the practical value of these approaches, with quantified improvements in system accuracy, operational efficiency, and compliance capabilities. As organizations pursue digital transformation and seek to leverage AI for mission-critical decision support, ontology-grounded reasoning systems position themselves as essential infrastructure enabling the next generation of intelligent, accountable, and semantically coherent AI systems.
References
[1] 	T. Schneider and M. Šimkus, “Special issue on ontologies and data management: Part i,” KI - Künstliche Intelligenz, Sept. 2020, doi: 10.1007/s13218-020-00682-7.
[2] 	Ms. S. Prathi, M. Fathima, S. Rumaiza, and D. Elavarasi, “Ontologies and artificial intelligence: Bridging semantic gaps in machine understanding,” 2025 7th International Conference on Innovative Data Communication Technologies and Application (ICIDCA), Oct. 2025, doi: 10.1109/ICIDCA66325.2025.11280578.
[3] 	W. Hieke, S. Schwöbel, and M. Smolka, “Involving cognitive science in model transformation for description logics,” Logic Journal of the IGPL, Aug. 2024, doi: 10.1093/jigpal/jzae088.
[4] 	V. Huzhva, “ONTOLOGICAL KNOWLEDGE MODELING IN THE DIGITAL TRANSFORMATION OF ACADEMIC INSTITUTIONS,” Intellect XXІ, 2025, doi: 10.32782/2415-8801/2025-2.12.
[5] 	O. Gorda, Y. Riabchun, R. Mazurenko, and V. Khrolenko, “Ontology-based analysis of neuralstem learning based on data integration,” 2025 IEEE 5th International Conference on Smart Information Systems and Technologies (SIST), May 2025, doi: 10.1109/SIST61657.2025.11139176.
[6] 	X. Yang, T. Wei, X. Li, and Z. Qian, “Ontology-based knowledge organization of bronze weapons from the shang and zhou dynasties,” Knowledge organization, Feb. 2026, doi: 10.31083/ko46922.
[7] 	L. Asprino and V. Presutti, “Observing LOD: Its knowledge domains and the varying behavior of ontologies across them,” IEEE Access, 2023, doi: 10.1109/ACCESS.2023.3250105.
[8] 	A. Katipelly and S. Thalary, “Semantic automation of basel III liquidity reporting: Utilizing ontological knowledge graphs for real-time regulatory compliance and auditability,” International Journal of Emerging Research in Engineering and Technology, 2024, doi: 10.63282/3050-922x.ijeret-v5i2p115.
[9] 	H. Sridharan, S. Sarkar, P. Paul, and E. Cheng, “Toward a hybrid ontology framework for semantic data understanding in AI-augmented data warehousing,” International Conference on Big Data Computing Service and Applications, July 2025, doi: 10.1109/BigDataService65758.2025.00033.
[10] 	O. Martinez and D. Sharma, “Semantic web technologies for knowledge graph construction and querying,” International Journal on Advanced Computer Theory and Engineering, Apr. 2025, doi: 10.65521/ijacte.v13i1.94.
[11] 	T. Nipurakis, S. Chatzinikolaou, G. Vassiliou, and N. Papadakis, “Spatial and temporal knowledge representation: Ontological foundations, semantic web standards,” Electronics, Apr. 2026, doi: 10.3390/electronics15081590.
[12] 	D. Herron, E. Jiménez-Ruiz, and T. Weyde, “On the potential of logic and reasoning in neurosymbolic systems using OWL-based knowledge graphs,” Neurosymbolic Artificial Intelligence, Apr. 2025, doi: 10.1177/29498732251320043.
[13] 	G. Xiao and J. Corman, “Ontology-mediated SPARQL query answering over knowledge graphs,” Big Data Research, Feb. 2021, doi: 10.1016/j.bdr.2020.100177.
[14] 	S. M. B. Silalahi, R. Indrajit, and T. Mantoro, “Development of a semantic-web enabled enterprise architecture framework for curriculum governance,” International Conference Computing Engineering and Design, Nov. 2025, doi: 10.1109/ICCED68324.2025.11324880.
[15] 	M. Ali, Z. Taha, and M. M. Morsey, “Ontology-grounded knowledge graphs for mitigating hallucinations in large language models for clinical question answering,” Journal of Biomedical Informatics, Jan. 2026, doi: 10.1016/j.jbi.2026.104993.
[16] 	U. Das, K. Atmakuri, D. H. Ho, C. Lee, and Y. Lee, “Clinical knowledge graph construction and evaluation with multi-LLMs via retrieval-augmented generation,” arXiv.org, Jan. 2026, doi: 10.48550/arXiv.2601.01844.
[17] 	D. H. Ho, U. Das, R. Ho, and Y. Lee, “Leveraging multi-agent systems and large language models for diabetes knowledge graphs,” BigData Congress [Services Society], Dec. 2024, doi: 10.1109/BigData62323.2024.10825608.
[18] 	N. Lazzari, V. Presutti, and A. Vergari, “To neuro-symbolic classification and beyond by compiling description logic ontologies to probabilistic circuits,” arXiv.org, Jan. 2026, doi: 10.48550/arXiv.2601.14894.