Introduction to ontologies: The backbone of semantic understanding

The text you are about to read was generated in conjunction with Google NotebookLM, through my research and studies in Ontologies to generate better projects related to SEO, information systems, the creation of Artificial Intelligence agents, and other related subjects.

It was generated through answers to my questions and organized to provide a general overview of what was published between 2022 and 2025 on this intriguing subject, which has become fundamental to all areas involved with digital web .

Formal definition and strategic importance

In the field of Artificial Intelligence ( AI ), an ontology is defined as a formal and explicit representation of knowledge, manifested as a set of concepts, entities , properties, and the relationships established between them within a specific domain. Fundamentally, an ontology articulates "what exists" in that domain and how its components interconnect, thus serving as the semantic backbone ^for intelligent systems. This ability to structure knowledge in a way that is understandable by machines is what positions ontologies as crucial elements in the pursuit of more robust and contextualized AI.

The relevance of ontologies has grown exponentially, driven by the proliferation of knowledge graphs data generated daily. Recent publications, such as the 2024 “Ontology Engineering” guide, emphasize that ontologies offer a rich and semantically grounded “schema” for the KGs that underpin these technologies. More than that, they provide the essential terminological and semantic foundation for achieving dramatic improvements in the outcomes and effectiveness of ML and NLP applications.<sup> ²

The increasing complexity and demand for greater context awareness in modern AI systems¹ ^make the structured representation of knowledge, as provided by ontologies, not only useful but indispensable. Without this formal semantic backbone, AI systems would face considerable difficulties in handling ambiguities, performing complex logical inferences, and providing transparent explanations of their decisions and behaviors. Ontologies, by improving semantic understanding in NLP systems, enabling reasoning through the definition of rules and logic, supporting the explainability of AI decisions, driving intelligent search, and being essential for interoperability¹ ^, transcend the concept of mere descriptive models. They are configured as active components that enhance the performance and comprehensibility of AI systems, becoming a practical necessity for the development and understanding of sophisticated, next-generation AI solutions.

Key components of an ontology

For an ontology to fulfill its role as a formal representation of knowledge, it is constructed from several key interrelated components. Understanding these building blocks is fundamental for anyone wishing to understand or work with ontologies. The main components include ¹:

• Classes: Represent categories, concepts, or types of entities in the domain. They function as abstractions that group individuals with similar characteristics. For example, in a transportation domain, Vehicle would be a class; in a biological domain, Animal and Person would be classes.
• Individuals or Instances: These are the specific occurrences or concrete examples of classes. If Vehicle is a class, a specific Tesla Model 3 would be an individual. Similarly, Tiger could be an individual of the Animal class, and Alice an individual of the Person class.
• Properties or Attributes and Relations: These describe the characteristics of classes and individuals, as well as the ways in which they relate to each other. Properties can be attributes that describe an individual (e.g., a vehicle has wheels, a person has height) or relations that connect two or more individuals (e.g., a person is a friend of another person, a vehicle is manufactured by a company).
• Axioms: These are formal statements or rules that are considered true within the domain of the ontology. Axioms restrict the meaning of terms and define the logic of the domain. An example of an axiom could be "All humans are mammals," which establishes a fundamental truth about the relationship between the classes Human and Mammal.
• Hierarchies: Classes are often organized into taxonomic hierarchies using subclass-superclass relationships (also known as "is-a" or "type-of" relationships). For example, the Car class can be a subclass of the Vehicle class (denoted as Car⊆Veıˊculo), indicating that every car is a type of vehicle and inherits properties from Vehicle.

The table below summarizes these fundamental components, offering a quick reference to consolidate an understanding of their roles in the structure of an ontology.

Table 1: Fundamental Components of an Ontology

These components, when combined logically and consistently, allow an ontology to provide a rich and formal specification of a domain of knowledge.

The crucial role of ontologies in modern AI.

In the landscape of modern Artificial Intelligence, especially in the period between 2022 and 2025, ontologies play an increasingly crucial and multifaceted role. They are instrumental in enhancing semantic understanding in Natural Language Processing (NLP) systems, allowing machines to interpret the meaning and context of human language with greater accuracy. By defining rules and formal logic, ontologies empower AI systems with reasoning abilities, enabling them to derive new knowledge from existing information. Furthermore, they are fundamental to supporting explainability (XAI) in AI decisions, providing a comprehensible trace of the logic used. Ontologies also drive intelligent search, such as in search engines , which go beyond keyword matching to understand user intent. Finally, they are essential for ensuring interoperability in complex systems, such as multi-agent systems and those operating across multiple domains, facilitating consistent communication and information exchange. ¹ In short, ontologies help machines understand the context , and not just the raw data with which they operate.

One particularly promising area is the integration of ontologies with Large Language Models (LLMs) and Generative AI (GenAI) tools. A perspective for 2025 indicates that incorporating domain ontologies into the prompting and memory systems of LLMs will bring significant benefits. This is expected to result in more accurate and relevant outputs in specific domains, better reasoning in multi-turn interactions, and greater factual grounding of the information generated by LLMs. In this scenario, ontologies could function as the structured and verified memory for LLMs.<sup> ¹

This synergy is particularly important considering a known weakness of LLMs: the tendency towards “hallucinations,” that is, the generation of information that seems plausible but is factually incorrect or contextually inappropriate. ^ontologies can act as an anchoring or reference mechanism for LLMs. By providing a well-defined factual and relational base, ontologies can guide the LLM generation process, restricting its outputs to more accurate and contextually appropriate information, especially in specialized domains that require high fidelity. This collaboration between the generative flexibility of LLMs and the semantic rigor of ontologies points to a future where Generative AI applications can become significantly more reliable and trustworthy. For those learning about ontologies, this signals a trajectory where LLM-based systems may increasingly rely on well-curated domain ontologies to reach their full potential.

Distinction and relationship between ontologies and Knowledge Graphs (KGs)

The terms "ontology" and "knowledge graph" ( KG ) are frequently used in the same context and sometimes interchangeably, which can lead to confusion, especially for those beginning their studies in the field. Although closely related, they refer to distinct but complementary concepts.

An ontology can be understood as the scheme or structure that defines a domain. It establishes the rules, classes (concepts), properties (relations), and axioms that govern that domain. One can think of ontology as a framework , generally more static, that describes the types of things that exist and how they can relate to each other. It is the “grammar” that defines the language for talking about a domain. ^¹

On the other hand, a knowledge graph is an instantiation of this structure with concrete data. It contains not only the structure (often provided by an ontology), but also the individuals (specific instances of the classes) and the factual relationships between them. Knowledge graphs are dynamic and populated with data, representing real knowledge about entities and their connections.<sup> ¹

The table below illustrates the main differences and the relationship between ontologies and knowledge graphs:

Table 2: Comparison: Ontologies vs. Knowledge Graphs

Source: Based on.¹

Despite their distinctions, ontologies and knowledge bases (KGs) work together to power a wide range of intelligent applications. The ontology provides the semantic model and inference rules, while the KG stores and organizes factual data according to that model. Together, they are the driving force behind systems like Google's search engine, virtual assistants like Siri and Alexa, and advanced AI-based chatbots. ^{Understanding} this distinction is fundamental to appreciating how knowledge is modeled, managed, and used in contemporary AI systems.

Types of Ontologies and their strategic roles

Ontologies can be classified in various ways, depending on their level of generality, scope, and purpose. Understanding the different types of ontologies is crucial for selecting or developing the most appropriate knowledge structure for a given application.

Common classifications: Dominant, Superior, and Hybrid

Three main types of ontologies are commonly discussed in the literature and in practice. ⁴:

• Domain Ontology: This type of ontology is designed to represent knowledge belonging to a specific field or area of ​​the world. For example, a domain ontology can be developed for biology, medicine, politics, software engineering, or any other specialized field. The main characteristic of a domain ontology is that it models definitions of terms that are specific and meaningful within that particular domain. An illustrative example is the word "card": an ontology about the domain of poker would model the meaning of "card" as "playing card," while an ontology about the domain of computer hardware would model the meanings of "punched card" and "video card. ^"
• Upper Ontology (or Foundation Ontology): In contrast to domain ontologies, a higher ontology models concepts, relationships, and objects that are very general and common, applicable to a wide range of domain ontologies. It aims to provide a framework of basic and universal concepts, such as time, space, process, object, and event, that can be reused and specialized by different domain ontologies. Higher ontologies generally employ a central glossary that covers terms and object descriptions in a generic way. Notable examples of higher ontologies include Basic Formal Ontology (BFO), Descriptive Ontology for Linguistic and Cognitive Engineering (DOLCE), and Suggested Upper Merged Ontology (SUMO) ^.
• Hybrid Ontology: ^ontologies . It seeks to integrate the general concepts of a top-level ontology with the specific terms of a particular domain. An example cited is the Gellish ontology, which incorporates elements of both types.

The interaction between these types of ontologies is fundamental for scalable knowledge management. Domain ontologies, when developed independently, often result in semantic incompatibilities due to different perceptions of the domain, languages, or purposes of use. Merging these distinct ontologies can be a manual, time-consuming, and expensive process.^⁴ However, using a common top-level ontology as a basis for developing multiple domain ontologies can significantly mitigate these challenges. By providing a standardized set of fundamental concepts and relationships, top-level ontologies facilitate integration and interoperability between domain ontologies that share this same base, reducing the manual merging effort.^⁴ This suggests that the development and adoption of robust top-level ontologies are critical to achieving broader semantic interoperability and knowledge reuse across different domains and applications. Hybrid ontologies, in turn, represent a pragmatic approach to leverage the strengths of both worlds, combining the generality of top-level ontologies with the necessary specificity of domain ontologies.

Core Ontologies and their relevance

In addition to the broader domain and higher-level classifications, the concept of core ontology is also relevant, especially in contexts of complex and integrated modeling. A core ontology is designed to capture the fundamental concepts and relationships that are central to a particular area of ​​interest, which may be broader than a specific domain but not as universal as a higher-level ontology.

A prominent and recent example is the Ontology of Descriptions and Observations for Integrated Modelling (ODO-IM) . This core ontology was developed to capture scientific observations and the descriptions related to those observations, with the goal of integrating diverse scientific assets. ^{ODO -} IM is a central component of k.LAB, an integrated modeling software, and relies on the Provenance Ontology (PROV-O) to track the origins and changes of information, which is crucial for representing scientific knowledge.

The ODO-IM framework is organized around three main categories (or “backbone”): Description (capturing the activities that produce scientific artifacts from contextualized observations), Observable (the concepts that are the object of a scientific description and provide the basic semantics for the resulting observation), and Predicate (linguistic constructs that can be combined with observables to constrain their meaning). ^{Within scientist} or a sensor) and an observation target (e.g., a physical object or an event).

Core ontologies, such as ODO-IM, often arise when existing higher-level ontologies are considered too generic or when their ontological commitments (the fundamental assumptions about the nature of reality that they embody) do not align with the specific requirements or “scientific vision” of a particular domain or platform. For example, ODO-IM makes specific design choices, such as not including semantics for units of measurement, because its primary applications in the k.LAB software (currently focused on social and environmental sustainability) do not require them. It also remains agnostic regarding certain philosophical distinctions that are integral to foundational ontologies like BFO. ⁵ This demonstrates that core ontologies seek a balance, providing fundamental concepts that are more specific than those of a higher-level ontology, but still general enough to cover a complex area of ​​interest without becoming as granular as highly specialized domain ontologies. They represent a pragmatic approach to ontology engineering, where developers choose to create or adapt ontologies that best fit their complex and immediate needs, rather than trying to adhere to a universal foundational model if it is not practical or suitable.

Considerations for choosing the type of ontology

The decision about which type of ontology to develop or use is not trivial and fundamentally depends on the purpose of the ontology and the context of its application. Although the principles outlined in the “Ontology Development 101” guide ^publication (circa 2000), the initial step of determining the domain and scope of the ontology remains a timeless and crucial driver for this choice.

The main factors to consider include:

• The domain that the ontology will cover: How broad or specific is the field of knowledge? A very narrow domain may benefit from a detailed domain ontology, while the need to integrate knowledge from multiple domains may point to the use of a higher-level ontology as a foundation.
• The intended use of the ontology: Will the ontology be used for data annotation, automated reasoning, information retrieval, interoperability between systems, or to facilitate human communication? Different uses may require different levels of formality and expressiveness.
• The types of questions the ontology should answer (competency questions): As suggested in ⁶ , formulating “competency questions” helps define the level of detail and the types of concepts and relationships that the ontology needs to capture. If the questions are very niche-specific, a domain ontology is likely. If the questions encompass fundamental concepts applicable to many areas, a higher-level ontology may be more relevant.
• Who will use and maintain the ontology: The complexity of the ontology and the tools required for its use and maintenance must be compatible with the technical knowledge of the users and maintainers.

There is no "one-size-fits-all" type of ontology. The choice is a strategic decision that involves balancing the need for specificity to solve problems in a particular domain with the benefits of generality for reusability and interoperability. For example, a project aiming to integrate diverse datasets in a large company might benefit from adopting a superior ontology as a common backbone, complemented by domain ontologies for specific business areas. In contrast, a research project focused on a very particular scientific problem might choose to develop a highly detailed and specialized domain ontology, possibly extending a relevant core ontology. The key is to align the ontology type and design with the application's objectives, ensuring it delivers the necessary semantic value effectively and efficiently.

Ontology engineering: from conception to practical implementation.

Ontology Engineering (OE) is the discipline that deals with the principles, methodologies, and tools for the development, lifecycle, and maintenance of ontologies. It is a field that combines aspects of knowledge representation, formal logic, linguistics , and domain expertise.

Fundamental principles and methodologies of development

Ontology development, while evolving with new tools and techniques, is still based on fundamental principles that guide the creation of robust and useful knowledge representations. The guide “Ontology Development 101”^⁶ , despite its original publication date, establishes some of these pillars that remain relevant:

1. There is no single correct way to model a domain: There will always be viable alternatives, and the best solution depends on the intended application and anticipated extensions.
2. Ontology development is an inherently iterative process: An ontology is rarely perfect on the first try; it evolves through cycles of design, evaluation, and refinement.
3. The concepts in ontology must closely reflect the objects (physical or logical) and relations in the domain of interest: They generally correspond to nouns (objects) and verbs (relations) in the sentences that describe the domain.

The inherent complexity of ontology engineering (OE) has led to the emergence of several formal methodologies. These methodologies aim to provide structured frameworks to assist ontology engineers in navigating the complexities of knowledge modeling. ^The need for these structured approaches is reinforced by the nature of OE itself, which is consistently described as a complex task, requiring not only technical knowledge of representation languages ​​such as OWL ^crucially , a deep understanding of the domain to be modeled.

Even with advancements in tools and accumulated experience in the field, ontology engineering remains a significant challenge. It is frequently characterized as a process that is “complex, time-consuming, and error-prone, even for experienced ontology engineers.” ³ The manual tasks of creating, curating, and validating ontological elements are described as “cognitively costly.” ³ This persistent difficulty is one of the main motivators for research and development of approaches that can automate or assist parts of the ontology engineering process, such as the use of Large Language Models (LLMs), which will be discussed later. The complexity lies not only in the technical aspects but also in the conceptual challenges of accurately and consistently capturing the semantics of a domain, and often in the social challenges of reaching consensus among multiple domain experts.

The ontology lifecycle: essential steps

The development of an ontology follows a lifecycle that spans from initial conception to its application and ongoing maintenance. Although different methodologies may propose variations, a set of essential steps is commonly recognized. Adapting the guidelines of ⁶ to a contemporary context, and supplementing them with more recent observations, the lifecycle can be outlined as follows:

1. Determining the Domain and Scope: This is the fundamental step. It involves clearly defining which area of ​​knowledge the ontology will cover, what its main purpose will be, what types of questions it should be able to answer (the "competency questions"), and who its users and maintainers will be.
2. Consider Reusing Existing Ontologies: Before starting to build a new ontology from scratch, it is crucial to investigate whether ontologies (or parts thereof) already exist that can be reused, adapted, or extended. This can save significant time and effort and promote interoperability.
3. Enumerate Important Terms: Compile a comprehensive list of all relevant terms in the domain that the ontology will represent. At this stage, the focus is on breadth, without worrying excessively about overlaps or the exact nature of each term (whether it is a class, property, etc.).
4. Defining Classes and Class Hierarchy: Organize the identified terms into classes (concepts) and structure them in a taxonomic hierarchy (subclass-superclass relationships). This can be done through top-down (from general to specific), bottom-up (from specific to general) approaches, or a combination of both.
5. Defining Class Properties (Slots): Identify and define the attributes of the classes and the relationships between them. Properties are inherited by subclasses in the hierarchy.
6. Defining Property Facets (Slots): Specify the characteristics of properties, such as the type of value they can take (e.g., string, number, boolean, instance of another class), the cardinality (how many values ​​the property can have – one or multiple), and the domain (the class to which the property applies) and codomain (the type of value the property can have).
7. Create Individuals (Instances): Populate the ontology with specific instances of the defined classes, filling in the values ​​of their properties.

It is important to emphasize that this process is typically iterative. The ontology is developed, evaluated, refined, and expanded in multiple cycles. Furthermore, the lifecycle does not end with the initial creation. The maintenance is crucial, as domains evolve, new knowledge emerges, and application requirements change. Recent research, such as that observed in the field of Ontology-Based Safety Management (ObSM) in the construction industry, indicates a growing trend of focusing on the maintenance phase of the ontological lifecycle. ^⁷ Similarly, ontologies such as Artificial Intelligence Ontology (AIO) are designed to be “dynamically updated” to reflect rapid advancements in their domain.^⁸ This underscores that an ontology is a “living” artifact, requiring continuous attention to maintain its relevance and accuracy over time. Feedback from the application and integration phases frequently informs and guides subsequent development and maintenance activities.

Standard tools and languages

Ontology engineering is supported by an ecosystem of standardized tools and languages ​​that facilitate the creation, editing, visualization, querying, and reasoning about ontologies. Familiarity with these resources is essential for anyone wishing to engage in the practice of OE.

Standard languages ​​provide the syntax and formal semantics for expressing ontological knowledge in a way that is interpretable by both humans and machines. The most prominent of these is the Web Ontology Language (OWL) ^W3C ) that has become the de facto standard for creating ontologies in the Semantic Web and many other applications. OWL is based on Description Logics, which gives it a robust formal foundation and allows for different levels of expressiveness (profiles such as OWL DL, OWL Lite, OWL 2 QL, EL, RL). OWL, in turn, generally uses the Resource Description Framework (RDF) for serialization.

To work with these languages, a variety of tools have been developed:

• Ontology Editors:
• Protégé: Developed by Stanford University, it is the most popular and widely used open-source ontology editor. It offers a rich graphical interface for creating and editing ontologies in OWL and other formats, and supports plugins for reasoning tools and other functionalities ^.
• Frameworks and APIs:

• Apache Jena: A robust open-source Java framework for building Semantic Web and linked data applications. It provides APIs for reading, writing, and querying RDF and OWL data, and includes an inference engine ^.
• RDFLib: A Python library for working with RDF, allowing you to parse, serialize, query (using SPARQL), and manipulate RDF graphs. ¹
• Ontology Access Kit (OAK): A more recent toolkit focused on programmatic access to ontologies. It offers Python examples and a command-line interface (CLI), supporting various backends such as Triplestores, OBO Graphs, OWL, and SQLite files. ^{It performing} enrichment and overrepresentation analysis, and even integrating with LLMs.
• Query Languages:
• SPARQL (SPARQL Protocol and RDF Query Language): The standard query language for RDF data, allowing you to extract information from knowledge graphs and ontologies.

The table below summarizes some of the main tools and languages ​​used in ontology engineering:

Table 3: Tools and Languages ​​for Ontology Engineering

The choice of tools and languages ​​will depend on the specific project requirements, the team's familiarity, and the existing infrastructure. However, adherence to standards such as OWL and RDF is crucial to ensure the interoperability and longevity of the developed ontologies.

The rise of Large Language Models (LLMs) in ontology engineering

One of the most significant and recent trends (2022-2025) in ontology engineering is the exploration and application of Large Language Models (LLMs) to assist and potentially automate various tasks in the ontological lifecycle. Given the complex and often time-consuming nature of manual ontology engineering, LLMs emerge as a technology to accelerate the process and democratize access to ontology creation.

Recent research, such as that presented in March 2025 ^, investigates the potential of LLMs to generate effective ontology drafts in OWL format directly from ontological requirements described in natural language, such as user stories and competency questions. Notably, these studies indicate that certain LLM models, when properly guided, can even outperform novice ontology engineers in terms of modeling ability. ^Another study, from February 2025, focused on fine-tuning LLMs such as GPT-4 and Mistral 7B, using fundamental OE texts as a basis for creating training datasets. The results showed that the tuned GPT-4 demonstrated superior accuracy and greater adherence to ontological syntax, albeit with higher computational costs, while Mistral 7B excelled in speed and cost-effectiveness, but with challenges in syntactic accuracy for domain-specific tasks. A crucial point highlighted was the need for domain-specific datasets to significantly improve the performance of LLMs in these tasks. ¹²

Despite the enthusiasm, the application of LLMs in EO is not without challenges. The same studies that point to the potential also highlight “common errors and variability in the quality of results.” ^Issues such as “hallucinations” (generation of incorrect or fabricated information), the difficulty in ensuring logical consistency, and the need for a multidimensional evaluation (going beyond simple automated metrics) are active concerns. ^To prompting techniques (careful formulations of the instructions given to the LLM) are being explored. Among them, the Chain of Thought (CoT), Metacognitive Prompting (MP), and prompting ^, . These aim to guide the LLM through more structured steps to improve the quality of the generated ontology.

The emerging role of LLMs in OE appears to be that of accelerators and enhancers of human capabilities, and not (at least at the current stage of technology) complete replacements for ontology engineers. LLMs can “provide effective drafts of OWL ontologies” ³ and “significantly reduce the time and effort involved in manually building ontologies.” ¹² However, human oversight remains crucial for the validation, refinement, and quality assurance and correctness of the generated ontologies, especially to handle complex domain nuances and ensure logical consistency. ⁹ The role of the ontology engineer is therefore evolving: from a purely manual creator to a curator and guide of AI-assisted processes, involving the formulation of prompts , the creation of datasets for fine-tuning , the critical evaluation of LLM outputs, and the integration of these outputs into a cohesive and correct final ontology. Future workflows in ontology engineering will likely be hybrid, combining human expertise with the processing and generation capabilities of LLMs. For those aspiring to work in this field, this implies the need to develop competencies in both traditional ontology engineering principles and the effective use of these new AI tools.

The Importance of FAIR Principles for Ontologies

In recent years, the FAIR principles – Findable , Accessible , Interoperable , and Reusable – have become a benchmark for managing and sharing research data. Recently, the application of these principles has also been extended to ontologies themselves, recognizing them as valuable knowledge artifacts that greatly benefit from increased findability, accessibility, interoperability, and reusability.

A tutorial given by Luiz Bonino at the Ontobras 2022 event, for example, specifically addressed the intersection between FAIR principles and the ontology engineering process, aiming to help participants understand how to transform their ontologies into FAIR artifacts. ^{more The} ODO-IM ontology itself, mentioned earlier, was developed with a focus on modularity and reusability, aligning with FAIR principles.

Applying FAIR principles to ontologies acts as a quality standard and a facilitator for more robust and collaborative ontology ecosystems. When an ontology is:

• Findable: It can be discovered by other researchers and systems through rich metadata and persistent identifiers, increasing its visibility and potential for use.
• Accessible: Its terms, definitions, and structure can be accessed and retrieved by humans and machines through standardized protocols, allowing for its inspection and use.
• Interoperable: It can be combined and used in conjunction with other ontologies and data sources, using standard languages ​​and formats (such as OWL and RDF), which is crucial for large-scale knowledge integration.
• Reusable: Its components (classes, properties, axioms) can be reused in different contexts and for different purposes, with clear licenses and adequate documentation, saving effort and promoting consistency.

Adopting FAIR principles in ontology engineering is not just a good practice, but a strategy to increase the impact and value of developed ontologies. FAIR ontologies are more likely to be adopted by the community, integrated into diverse applications, and maintained over time. This fosters a richer and more interconnected semantic ecosystem where knowledge can be shared and combined in more effective ways. For ontology developers and users, choosing or developing ontologies that adhere to FAIR principles means contributing to, and benefiting from, a more collaborative and sustainable knowledge landscape.

Advanced Ontology Use Cases (2022-2025)

Ontologies have moved beyond being merely a topic of academic research and have become integral components in a myriad of advanced applications, especially in the period from 2022 to 2025. Their ability to provide a formal and shared representation of knowledge has been exploited in various sectors to solve complex problems.

Ontologies in Artificial Intelligence projects

The synergy between ontologies and Artificial Intelligence is profound and multifaceted. As AI systems become more complex and are required to operate with greater context awareness, the need for structured knowledge, which ontologies provide, becomes paramount.¹

Enhancing Knowledge Representation, Reasoning, and Explainability

Ontologies are crucial for enhancing knowledge representation in AI systems, allowing concepts and their relationships to be defined explicitly and unambiguously. This, in turn, improves semantic understanding and empowers systems with more sophisticated reasoning abilities. A particularly important development is the role of ontologies in promoting Explainable AI (XAI) ^{2025 which} combines machine learning approaches with symbolic reasoning).

A November 2023 article, titled “On the Multiple Roles of Ontologies in Explainable AI” ¹⁵ , delves deeper into this discussion, identifying how ontologies contribute to XAI through:

• Reference Modeling: Providing a shared conceptual model of the domain that can be used as a basis for generating explanations.
• Common Sense Reasoning: Incorporating common sense knowledge that can make explanations more intuitive and understandable for humans.
• Knowledge Refinement and Complexity Management: Helping to structure and simplify complex knowledge, making it more manageable for explanation purposes.

The ability of ontologies to formalize concepts and relationships in a way that is understandable by humans makes them a vital bridge to making AI systems more transparent and trustworthy. As AI models, especially deep learning models, increasingly operate as “black boxes,” the demand for interpretability and transparency grows. Ontologies offer a semantic layer that can be used to translate the complex behaviors of these models into terms and reasoning steps that humans can understand, audit, and consequently trust. The integration of ontologies is becoming a key strategy for the development of XAI systems, addressing one of the main societal concerns about the impact and reliability of AI. For developers, this means that ontologies are not just tools for knowledge representation, but also essential components for building a more responsible AI aligned with human values.

Synergy between ontologies and LLMs for more accurate and contextual AI.

The relationship between ontologies and Large Language Models (LLMs) is increasingly seen as symbiotic. As discussed earlier (Section 1.3), incorporating domain ontologies into LLMs has the potential to significantly improve the accuracy of their outputs, enhance their multi-turn reasoning capabilities, and provide a more solid factual foundation for the information they generate. ^mitigate problems such as LLM “hallucinations,” where they can generate incorrect or nonsensical content .

On the other hand, LLMs are also being used to assist in the development and maintenance of ontologies. The case of the Artificial Intelligence Ontology (AIO), which will be detailed below, is an example where LLMs were used in the curation process. ⁸ This bidirectional relationship suggests a virtuous cycle: LLMs help build better ontologies more quickly, and these ontologies, in turn, help make LLMs more accurate, reliable, and contextually aware.

Case Study: An Artificial Intelligence Ontology (AIO)

A paradigmatic example of the application of ontologies in the field of AI itself is the Artificial Intelligence Ontology (AIO) , detailed in an April 2024 publication. ^⁸ The AIO is a formal systematization of the concepts, methodologies, and their interrelationships in the vast and dynamic domain of Artificial Intelligence. Its development combined specialized manual curation with the assistance of LLMs, reflecting modern ontology engineering practices. The main objective of the AIO is to standardize terminology and concepts within the field of AI, serving as a valuable resource for researchers, developers, and educators.<sup> ⁸

The AIO framework is organized around six high-level branches: Networks , Layers , Functions , LLMs , Preprocessing , and, crucially, Bias . The inclusion of “Bias” as a high-level class underscores the ontology’s commitment to encompassing not only the technical aspects but also the ethical and social considerations inherent in AI technologies. ⁸

The usefulness of AIO has already been demonstrated through practical applications, such as the annotation of data on AI methods in a catalog of research publications and its integration into BioPortal, a repository of biomedical ontologies, which highlights its potential for interdisciplinary research. ⁸ AIO serves as an excellent case study of a recent ontology, developed for a complex and rapidly evolving domain, using contemporary development approaches (LLM assistance) and aiming for high-impact applications (standardization, annotation, and fostering ethical discussions).

Sectoral Applications (e.g., Healthcare, Finance, Construction Industry)

The application of ontologies transcends the field of AI research and extends to various industrial and service sectors, where they provide tangible value by solving domain-specific problems.

• Healthcare: Ontologies are widely used in the medical field. For example, diagnostic models can be built using standardized medical ontologies such as SNOMED CT and the Unified Medical Language System (UMLS). ¹ An analysis of the best articles in Knowledge Management and Representation (KRM) in medicine in 2022 revealed a considerable focus on the creation of ontologies and knowledge graphs for the sector. ¹⁶
• Finance: In the financial sector, ontologies are applied to tasks such as risk categorization and fraud detection. The topic “Ontologies in Finance and Manufacturing” is scheduled for discussion at the onto:Nexus Forum 2025, indicating ongoing activity in this ^{area .}
• Construction Industry: This sector has seen an increase in the use of ontologies to improve information management. Applications include the management of various types of safety records (SM) ⁷ (published in January 2023) and the development of specific ontological models, such as the Safety and Health Exchange (SHE), for safety risk management in the design and planning phases ¹⁸ (published in 2022). Another example is the development of an ontology for rework management, aiming to support managerial decision-making and continuous improvement ¹⁹ (published in 2023).

These examples demonstrate that the practical adoption of ontologies is often driven by their ability to address specific domain needs. Ontologies such as SNOMED in healthcare or SHE in construction safety address the particular terminologies and challenges of their respective fields, resulting in improvements in data management, decision-making, and interoperability. The usefulness of an ontology is therefore directly proportional to its ability to accurately capture the semantics of the target domain and solve its specific problems. While generic ontologies can provide a foundation, it is the domain-specific extensions or dedicated ontologies that generally generate the most significant real-world impact.

Construction of semantic Information Retrieval tools

Information retrieval (IR) is an area that has benefited enormously from the application of ontologies, allowing the transition from systems based on simple keyword matching to truly semantic search tools. Ontologies drive intelligent search, enabling search engines to understand the intent behind user queries and the meaning of document content.¹

A comprehensive review on Semantic Information Retrieval and Ontology Engineering, published in July 2023 ^, highlights how ontological reasoning offers a formal, flexible, and scalable framework for knowledge representation, reasoning, and inference. This framework overcomes many of the limitations of traditional IR systems. The main techniques that utilize ontologies to enhance IR ^include :

• Query Expansion: Ontologies are used to expand user queries by adding synonymous, related, or hierarchically connected terms (more generic or more specific terms). This helps overcome vocabulary mismatch issues between the user's query and the documents, improving search scope.
• Semantic Annotation and Indexing: Documents can be annotated with concepts from an ontology, creating rich semantic metadata. This metadata allows for more precise indexing and matching based on meaning, rather than just the presence of keywords.
• Ontology Alignment and Integration: In scenarios where multiple sources of information or ontologies need to be considered, alignment and integration techniques are used to reconcile semantic differences and enable searches that encompass knowledge from diverse sources.

The application of ontologies is particularly valuable in retrieving information from complex digital collections, such as images. A study from July 2022 ²³ emphasizes that the use of ontologies is “very effective in improving the accessibility and accuracy of image retrieval,” helping to deal with ambiguities, doubts, and human metaphors that are difficult for machines to interpret based solely on visual content or limited textual metadata.

The challenge of information overload in the digital world makes it increasingly difficult to find relevant content efficiently. Traditional search engines, as noted in ²⁴ (although an older publication, the central problem persists), frequently struggle with query ambiguity and fail to provide adequate context for results. Ontologies address this issue by providing a “structured and standardized way of describing knowledge.” ²⁰ By enabling systems to understand the meaning behind queries and documents, ontologies are key to unlocking the real value contained in the vast ocean of digital information. Semantic search , powered by ontologies, is becoming essential for navigating complex information landscapes, offering users more relevant and accurate results and, for information systems developers, a critical component in building next-generation retrieval tools.

Data management and analysis powered by ontologies.

Data management and analysis are critical areas in all organizations, and ontologies are emerging as powerful tools to address persistent challenges such as data heterogeneity, interoperability between systems, and extracting meaningful insights from large volumes of information.

Integration of heterogeneous data and interoperability

One of the main benefits of ontologies in data management is their ability to facilitate the integration of data from heterogeneous sources and promote interoperability between different systems. ^Many organizations struggle with “data silos,” where valuable information is scattered across different databases, formats, and systems, each with its own schema and semantics. Ontologies offer a solution by providing a common vocabulary and a shared conceptual model that can be used to describe and relate this disparate data ^.

The ODO-IM ontology, for example, was explicitly designed to integrate various “scientific assets”. ⁵ Recent research on semantic data management, scheduled for publication in May 2025 ²⁶ , reinforces this view, highlighting that ontologies and knowledge graphs are powerful tools for increasing the usability and interpretation of data, facilitating data integration through approaches such as Ontology-Based Data Access (OBDA).

By mapping heterogeneous data sources to a common ontology, it is possible to overcome the semantic gaps between them. This allows data to be queried and analyzed in a unified way, as if residing in a single cohesive repository, even if they physically remain in their original locations. Ontologies, therefore, act as a "semantic glue" that unites disparate data silos, allowing a holistic view of information and, consequently, more informed decision-making and deeper insights.

Virtual Knowledge Graphs (VKGs) for unified data access

A specific architectural approach that uses ontologies for data integration is that of Virtual Knowledge Graphs (VKGs) . A 2022 tutorial on “Designing Virtual Knowledge Graphs with Ontop and Ontopic Studio” ^explains that VKGs work by exposing data from underlying sources through a flexible model – a knowledge graph whose vocabulary is defined by a domain ontology. Instead of materializing (copying and transforming) all the data into a new repository, VKGs maintain the virtual . This is achieved through mappings that specify how the data in the original sources correspond to the concepts and relationships in the ontology. Queries are made against the ontology, and the VKG system translates them into queries about the underlying data sources in real time.

Ontop is an example of a cutting-edge VKG system, compatible with Semantic Web standards such as RDF, OWL 2 QL, R2RML (for mappings), and SPARQL (for queries). ^{The OWL} language is particularly optimized for accessing large volumes of data, as it allows ontological queries to be efficiently rewritten into SQL queries (or other database query languages) on the underlying relational data. VKGs, therefore, offer a powerful and flexible way to achieve unified data access without the costs and complexity of large-scale data migration and synchronization.

4.3.3. Platforms for Semantic Data Management

Several platforms and tools are available or under development to support semantic data management based on ontology:

• k.LAB: It is an open-source web semantics software designed for integrated modeling, which uses the ODO-IM core ontology as its semantic reference. ⁵
• Ontop and Ontopic Studio: As mentioned, these are key tools for the design and implementation of Virtual Knowledge Graphs. ¹⁴
• Semantic Data Lakes: Research in semantic data management is also exploring the concept of “semantic data lakes,” which aim to integrate Big Data technologies with Semantic Web technologies (including ontologies and KGs) to manage and analyze large volumes of heterogeneous data in a more meaningful way. ²⁶

These platforms and approaches demonstrate the ongoing effort to make data management and analysis smarter, more efficient, and more meaning-driven, with ontologies playing a central role in this transformation.

Other advanced and innovative applications (2022-2025)

In addition to already established areas, ontologies continue to find new and innovative applications in diverse fields, reflecting their versatility and the growing recognition of their value.

An emerging field is the application of ontological thinking to interaction design and the critique of AI systems . A paper scheduled for the CHI '25 conference, entitled “Ontologies in Design: How Imagining a Tree Reveals Possibilities and Assumptions in Large Language Models,” ²⁸ proposes, according to its abstract, four orientations for considering ontologies in design: pluralism, groundedness, liveliness, and enactment. The aim is to use these orientations to analyze Large Language Models (LLMs), revealing potentialities and assumptions embedded in these systems. Although the full details of the paper were not available for this analysis beyond the abstract, the proposal itself indicates an innovative direction for the use of ontologies in the ethical and reflective evaluation and design of AI technologies.

The onto:Nexus Forum 2025 , an international forum on ontological modeling and analysis, also signals the breadth of contemporary applications. The event's agenda includes topics such as “Ontologies in Finance and Manufacturing,” “Space Modeling – A Suite of Ontologies for Astronautics and Astronomy,” and “Ontological Analysis of Building Codes in BIM (Building Information Modeling) Models.” ¹⁷ These themes illustrate how ontologies are being applied to structure knowledge and solve problems in domains as diverse as aerospace and civil construction regulation, all with a focus on current developments and discussions (2025).

The table below consolidates various ontology applications across multiple domains, focusing on the period from 2022 to 2025, to illustrate the breadth and impact of these technologies.

Table 4: Applications of Ontologies in Diverse Domains (2022-2025)

These examples demonstrate that ontologies are versatile tools, adaptable to a wide range of challenges and opportunities across multiple sectors, driving innovation and the pursuit of smarter and more understandable systems.

Current challenges and future opportunities in the use of ontologies.

Despite significant progress and the growing adoption of ontologies in various applications, the field still faces considerable challenges. At the same time, the continuous evolution of Artificial Intelligence, especially Large Language Models (LLMs), opens up new and promising opportunities.

Key obstacles in engineering, integration, and maintenance.

The journey from conceiving an ontology to its effective implementation and maintenance is fraught with obstacles that demand continuous attention and research.

• Complexity of Ontology Engineering (OE): ^already mentioned, OE remains an inherently complex, time-consuming, and error-prone process, even for experts. ^- date in the face of evolving domains is also a persistent challenge.
• Challenges with LLMs in EO: While LLMs offer potential to aid in EO, their use brings its own challenges. These include the occurrence of common errors in the generated drafts, variability in the quality of results, the need for robust multidimensional evaluation methods to validate the ontologies produced by LLMs, and the risk of “data leakage” (where the LLM may have been trained on the data it is being used to evaluate) ^.
• Obstacles in Semantic Information Retrieval: The application of ontologies in IR faces difficulties such as the acquisition and curation of ontological knowledge, the handling of ambiguity inherent in natural language, ensuring the scalability and adaptability of systems as data volume grows, the effective design and construction of the IR ontologies themselves, and overcoming semantic heterogeneity between different information sources and vocabularies. ²⁰
• Challenges in Semantic Data Management (especially for Big Data and Data Lakes): The application of ontologies in Big Data and data lake environments, while promising, faces significant barriers. These include the large initial investment required to create knowledge graphs and semantic mappings for a large number of heterogeneous datasets; the difficulty in evaluating the accuracy and quality of automatically generated semantic models; challenges in technical interoperability, especially with NoSQL data models and the need for federated query processing; the need for greater technical abstraction to make the tools usable by non-technical users; and ensuring applicability and performance in real-world Big Data scenarios. ²⁶

A recurring theme in many of these challenges is the trade-off between scalability and quality in ontology management . Manual ontology engineering, while capable of producing high-quality results, is slow, expensive, and difficult to scale. ^hand , automated methods, including those based on LLMs, promise greater scalability and speed, but often face problems related to quality, consistency, accuracy, and the difficulty of rigorously evaluating results. ^A significant and ongoing research direction is therefore to find the right balance between automated techniques for speed and scale, and human oversight and expertise to ensure quality, accuracy, and relevance. Hybrid approaches, combining the best of both worlds, are likely the most viable path forward, recognizing that there are no easy solutions and that the field is actively seeking ways to make ontology management both effective and efficient in the face of increasing demands.

Emerging opportunities with the evolution of AI, especially LLMs.

Despite the challenges, the future of ontologies is promising, largely due to the opportunities opened up by the continuous evolution of Artificial Intelligence, and in particular, Large Language Models (LLMs).

• LLMs as Tools to Support Ontology Engineering: LLMs have the potential to significantly reduce manual work for experienced ontology engineers and provide valuable assistance to newcomers to the field, lowering the barrier to entry for ontology creation. ³ Fine -tuning LLMs with domain-specific datasets from ontology engineering or particular application domains can further increase their usefulness and the quality of the generated ontologies. ¹²
• Synergy between Ontologies and Generative AI: The integration of ontologies with LLMs for Generative AI (GenAI) applications is an area of ​​great potential. By providing a structured and factual knowledge base, ontologies can help GenAI produce more accurate, contextually relevant, and well-founded outputs, mitigating problems such as the generation of incorrect information.^¹
• LLMs in Semantic Data Management: In the context of semantic data management, LLMs open opportunities to automate tasks such as describing data sources, initially creating domain ontologies from textual or semi-structured data, and discovering semantic relationships between different datasets. ²⁶

co-evolution between ontologies and LLMs is observed . On the one hand, LLMs are increasingly being used as tools to build and maintain ontologies more efficiently. ^other hand, ontologies are being proposed and used to improve the performance and reliability of the LLMs themselves, providing them with structured domain knowledge, reasoning capabilities, and a basis for more transparent explanations. ^This reciprocal relationship suggests that the future development of both fields will likely be increasingly intertwined. Advances in LLMs will lead to more sophisticated and accessible ontology engineering tools, while richer and more elaborate ontologies will enable more powerful, reliable, and intelligent LLM applications. This points to a future where AI systems will be built on a closer integration of symbolic (ontology-based) and sub-symbolic (LLM-based) approaches, leading to systems that combine the ability to learn from large volumes of data with the accuracy and interpretability of formal knowledge.

The importance of community collaboration and standardization.

The advancement and widespread adoption of ontologies depend not only on individual technical progress, but also on the strength and collaboration of the community of researchers and practitioners, as well as adherence to standards and best practices.

Initiatives such as the Artificial Intelligence Ontology (AIO), which is an open-source project that encourages community contributions ^, and forums like the onto:Nexus Forum, which aims to foster collaborations in the ontological modeling community ^, are examples of the importance of collaborative work. Events like Ontobras, with its tutorials on current topics such as FAIR principles for ontologies ^, also play a vital role in disseminating knowledge and strengthening the community.

The importance of standards such as OWL ¹ for ontology representation and the application of principles such as FAIR (Findable, Accessible, Interoperable, Reusable) ⁵ are fundamental to ensuring the interoperability and reusability of ontologies. Successful ontologies and initiatives often emphasize open standards, community collaboration, and resource sharing. The development of complex knowledge structures such as ontologies benefits immensely from the diversity of expertise and shared efforts that an active community can provide. Standards and principles provide the common ground necessary for such collaboration and to ensure that ontologies can be widely used, integrated, and understood by different systems and teams.

Progress in the field of ontologies is therefore catalyzed not only by isolated technical advances, but by the construction of a collaborative ecosystem. Openness promotes wider adoption, facilitates feedback, and drives continuous improvement, while standardization ensures that efforts are not fragmented and that ontologies can interact effectively. The future impact of ontologies will depend heavily on the vitality of this community and adherence to shared standards and best practices. For learners and professionals in the field, engaging with these communities, contributing to standards, and adopting principles of openness are key steps to staying up-to-date and contributing effectively to the advancement of the field.

Navigating the semantic future with ontologies

Recapitulation of the importance and versatility of ontologies.

Throughout this analysis, the multifaceted nature of ontologies was explored, from their fundamental components and development principles to their advanced applications in the 2022-2025 technological landscape. It became evident that ontologies are much more than mere data structures; they are formal and explicit representations of knowledge that serve as the semantic backbone for a vast range of intelligent systems. Their ability to define concepts, properties, and relationships in an unambiguous and machine-understandable way makes them indispensable for enhancing contextual understanding in Artificial Intelligence, for boosting the retrieval of truly semantic information, for facilitating the management and integration of heterogeneous data, and for driving innovation in domains as diverse as healthcare, finance, the construction industry, and scientific research. The distinction between ontologies and knowledge graphs, as well as the understanding of the different types of ontologies – domain, top-level, core, and hybrid – are crucial for appreciating the adaptability of these tools to different needs and scopes.

Perspectives on the evolution and adoption of ontologies

Looking to the future, the trajectory of ontologies seems intrinsically linked to the evolution of Artificial Intelligence itself. Synergistic integration with Large Language Models (LLMs) represents one of the most promising frontiers, where LLMs can help overcome the challenges of ontology engineering, while ontologies can provide LLMs with the factual foundation and semantic structure necessary for greater accuracy and reliability. The growing emphasis on FAIR principles for ontologies signals a movement towards a more open, collaborative, and interoperable knowledge ecosystem. The pursuit of greater automation in ontology engineering, driven by AI, will continue to be an area of ​​intense research and development, aiming to make the creation and maintenance of ontologies more accessible and efficient.

The expansion of ontologies into new domains and application types, such as interaction design and the analysis of AI systems from an ontological perspective, demonstrates the enduring relevance of structured thinking about knowledge. Challenges related to complexity, scalability, quality, and evaluation persist, but are accompanied by a vigorous effort from the scientific and industrial community to develop new methodologies, tools, and standards.

Ultimately, as highlighted at the beginning, in the era of increasingly context-driven and reasoning-oriented AI, ontologies have ceased to be an optional component and have become fundamental elements. ^“ understand” the world around them, allowing us to navigate the semantic future with greater clarity, precision, and intelligence.

References cited

1. Ontology in AI (2025 Guide): Structure, Semantics & Applications in …, accessed May 29, 2025, https://dev.to/bikashdaga/ontology-in-ai-2025-guide-structure-semantics-applications-in-knowledge-representation-44aa
2. Ontology Engineering (Synthesis Lectures on Data, Semantics, and Knowledge): 9783031794858 – Amazon.com, accessed May 29, 2025, https://www.amazon.com/Ontology-Engineering-Synthesis-Semantics-Knowledge/dp/3031794850
3. arXiv:2503.05388v1 [cs.AI] 7 Mar 2025, accessed May 29, 2025, https://arxiv.org/pdf/2503.05388?
4. Ontology (information science) – Wikipedia, accessed May 29, 2025, https://en.wikipedia.org/wiki/Ontology_(information_science)
5. Ontology of Descriptions and Observations for Integrated Modeling (ODO-IM) – Semantic Web Journal, accessed May 29, 2025, https://www.semantic-web-journal.net/system/files/swj3663.pdf
6. protege.stanford.edu, accessed May 29, 2025, https://protege.stanford.edu/publications/ontology_development/ontology101.pdf
7. A Review of Ontology-Based Safety Management in Construction – MDPI, accessed May 29, 2025, https://www.mdpi.com/2071-1050/15/1/413
8. [2404.03044] The Artificial Intelligence Ontology: LLM-assisted construction of AI concept hierarchies – arXiv, accessed May 29, 2025, https://arxiv.org/abs/2404.03044
9. The Artificial Intelligence Ontology: LLM-assisted – construction of AI concept hierarchies – arXiv, accessed May 29, 2025, https://arxiv.org/pdf/2404.03044?
10. arxiv.org, accessed May 29, 2025, https://arxiv.org/pdf/2404.03044
11. Ontology Access Kit (OAK) Documentation – GitHub Pages, accessed May 29, 2025, https://incatools.github.io/ontology-access-kit/
12. Fine-Tuning Large Language Models for Ontology Engineering: A …, accessed May 29, 2025, https://www.mdpi.com/2076-3417/15/4/2146
13. arxiv.org, accessed May 29, 2025, https://arxiv.org/pdf/2503.05388
14. Tutorials 2022 – ONTOBRAS, accessed May 29, 2025, https://www.inf.ufrgs.br/ontobras/en/tutorials-2022/
15. [2311.04778] On the Multiple Roles of Ontologies in Explainable AI – arXiv, accessed May 29, 2025, https://arxiv.org/abs/2311.04778
16. Knowledge Representation and Management 2022: Findings in Ontology Development and Applications – PMC, accessed May 29, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC10751114/
17. onto:Nexus Forum 2025 – openCAESAR, accessed May 29, 2025, https://www.opencaesar.io/events/onto-Nexus-Forum-2025
18. CONSTRUCTION SAFETY ONTOLOGY DEVELOPMENT AND ALIGNMENT WITH INDUSTRY FOUNDATION CLASSES (IFC), accessed May 29, 2025, https://www.itcon.org/papers/2022_05-ITcon-Farghaly.pdf
19. Full article: Curating a domain ontology for rework in construction: challenges and learnings from practice, accessed on May 29, 2025, https://www.tandfonline.com/doi/full/10.1080/09537287.2023.2223566
20. (PDF) Comprehensive Review on Semantic Information Retrieval and Ontology Engineering, accessed May 29, 2025, https://www.researchgate.net/publication/372625265_Comprehensive_Review_on_Semantic_Information_Retrieval_and_Ontology_Engineering
21. [2307.13427] Comprehensive Review on Semantic Information Retrieval and Ontology Engineering – arXiv, accessed May 29, 2025, https://arxiv.org/abs/2307.13427
22. arxiv.org, accessed May 29, 2025, https://arxiv.org/pdf/2307.13427
23. Application of Ontologies in Information Retrieval of Digital …, accessed May 29, 2025, https://jks.atu.ac.ir/article_13594.html?lang=en
24. [1012.1617] User Centered and Ontology Based Information Retrieval System for Life Sciences – arXiv, accessed May 29, 2025, https://arxiv.org/abs/1012.1617
25. (PDF) Ontologies: Principles, methods and applications – ResearchGate, accessed May 29, 2025, https://www.researchgate.net/publication/302937543_Ontologies_Principles_methods_and_applications
26. (PDF) A survey on semantic data management as an intersection of …, accessed May 29, 2025, https://www.researchgate.net/publication/380158769_A_survey_on_semantic_data_management_as_intersection_of_ontology-based_data_access_semantic_modeling_and_data_lakes
27. www.inf.ufrgs.br, accessed May 29, 2025, https://www.inf.ufrgs.br/ontobras/wp-content/uploads/2022/11/ontobras-2022-tutorial-vkgs.pdf
28. [2504.03029] Ontologies in Design: How Imagining a Tree Reveals Possibilities and Assumptions in Large Language Models – arXiv, accessed on May 29, 2025, https://arxiv.org/abs/2504.03029
29. Ontologies in Design: How Imagining a Tree Reveals Possibilities and Assumptions in Large Language Models – arXiv, accessed May 29, 2025, https://arxiv.org/html/2504.03029v1