Ontologies. A tool supporting AI and business

The term ‘ontology’ is a borrowing of the New Latin ontologia, which derives from the Greek ōn (ontos; being) and λόγος (logy; the study of). In philosophy, ontologies are seen as the study of being. They encompass the structure of reality, the concept of being, existence, modes of existence and its object, properties, causality, time, space, possibilities and necessities. Ontology in philosophy is sometimes equated with metaphysics. 

As the sciences of information technology and artificial intelligence developed, so the concept of ontology began to be used in the context of computers. In the nineteen sixties and seventies, pioneers in the field of AI like John McCarthy and Marvin Minsky started to create the first expert systems to need ways of formalising knowledge. What were known at the time as ‘frames’ and ‘semantic networks’ were used to that end and they became the origins of contemporary ontologies.

A breakthrough in the evolution of ontology in IT came with the development of knowledge representation languages like the Resource Description Framework (RDF) and Web Ontology Language (OWL). RDF, which was developed by W3C in the nineteen nineties, made it possible to model relationships between web resources, while OWL, introduced at the beginning of the twenty-first century, supplied tools for creating more advanced ontological structures. With OWL came the capability of defining the classes, properties, relationships and rules that allow machines to understand the significance of data and make inferences from them.

Ontologies in IT are formal representations of knowledge that describe concepts in a given field and the relationships between them. Their purpose is to organise and structure information in a way that is both comprehensible to humans and processable by computers. Ontologies constitute a particular kind of ‘skeleton’ of knowledge that enables computer systems to understand, classify and process information.

In IT, ontologies consist of the following elements: 

1.  classes (classifications), which represent concepts or categories like ‘vehicle’, ‘human being’ and ‘transaction’. Classes can be organised hierarchically, creating a tree or web of concepts with the more general concepts at the top and the more detailed ones lower down;

2.  properties (attributes), which define the features of characteristics of classes. A vehicle, for instance, can have properties like ‘colour’, ‘model’ and ‘year of manufacture’; 

3.  relations, which describe the connections between classes. For example, an ‘owner of the vehicle’ class could connect the ‘human being’ class with the ‘vehicle’ class.

4.  axioms, which describe the logic that applies to an ontology. An axiom might define every ‘vehicle’ as having to have one or more owners, for instance;

5.  individuals (instances), which are specific occurrences of classes. If the ‘vehicle’ class represents a general concept, then an individual, or instance, will represent a particular 2020 Ford Mustang belonging to a person named John Smith, for example.

Ontologies and artificial intelligence are closely connected because both these technologies have pivotal roles to play in the development of smart systems. Ontologies provide the formal, structural frames that permit AI systems to understand and interpret data in a more organised and comprehensible way.

AI is rooted in the capability of processing and analysing vast quantities of data. Ontologies make it possible to organise these data in a hierarchical structure that defines the relations between concepts, objects and their properties. The outcome is that AI has a better capability for understanding how various elements of the data are connected to each other, which leads to more precise results.

Medicine can serve as an illustration here. Medical ontologies assist AI systems in understanding relations between diseases, symptoms, medications and medical procedures. An AI system equipped with an ontology of that kind, for example, could diagnose patients better by proposing the most likely disease on the basis of the symptoms provided.

Ontology does more than just organise knowledge. It also expedites inference, which is to say, drawing logical conclusions on the basis of the information available. In AI systems, ontological inference makes it possible to deduce new facts on the basis of existing knowledge. This means that AI can, for example, automatically identify missing information, suggest alternative solutions to a problem or predict the consequences of particular actions.

In the AI context, ontologies are often used in advisory systems, decision support systems (DSSs)  and robots, where it is crucial for the system to understand context and have the capability of making decisions on that basis.

Ontologies are equally as critical to ensuring the interoperability of various AI systems. With their standardised terminologies and relations, ontologies enable ‘mutual understanding’ between AI systems, which is vital in complex environments like smart cities, where multiple systems have to work together.

Ontologies and machine learning are two different, but complementary, fields of artificial intelligence, which have the capability of working together to create systems that are more advanced and more effective. Let’s look at how they do that.

Ontologies provide formal models of knowledge representation that define the relations between concepts in a given field. In the context of machine learning, ontologies can be used to enrich and structure data, which is a key aspect of training models effectively.

Supervised learning is a category of machine learning where algorithms are trained with a technique using sets of labelled training data; in other words, the datasets contain inputs, referred to as features, and the correct corresponding outputs, known as labels. A model analyses the data in order to learn patterns and is then used to predict results for new, unknown data.

In supervised machine learning, ontologies can support the data labelling process. They define different classes and categories, making the labelling process more cohesive and more consistent with reality. They can also assist in detecting anomalies in data, which is useful when it comes to cleaning and preparing data for model training.

Machine learning frequently requires large quantities of data to reach a high level of accuracy. However, in some instances, such as medicine and law, data availability might be limited. Ontologies contain extensive domain knowledge and can thus fill in the gaps by supplying additional information that supports the algorithmic learning process.

Model interpretability is one of the challenges of machine learning, particularly as regards complex algorithms like neural networks. Ontologies can be helpful in explaining how a model has reached a given decision. This, in turn, means that the results will be clearer for users, which is especially important in crucial applications where understanding the decision-making process is critical.

Ontologies can also support the automation of tasks connected with machine learning, like feature selection and optimising hyperparameters. Clearly defined relations and dependencies in ontologies expedite a more effective exploration of a feature space and model configurations, which leads to the faster achievement of optimal results.

In advanced hybrid systems, where traditional symbolic methods like ontologies are combined with machine learning, ontologies can serve as a control or interpretative layer. In image recognition systems, for example, they can assist with categorising and clarifying results generated by neural networks and improve the quality and accuracy of the data interpretation at the same time.

Large language models (LLMs) like GPT-3 and GPT-4 are language models that are based on transformer architecture and have been trained on vast textual datasets. They have the capability of generating texts, translations and summaries and of responding to questions, often at a level close to human. The fact that they are trained on massive quantities of data means that they have an extensive knowledge base on a range of topics. Their understanding of context and the relationships between concepts, on the other hand, can be limited. Ontologies can be used to tackle this problem and improve LLM functioning by contributing a wider context and explaining relationships.

Although LLMs have vast quantities of data at their disposal, they are not always capable of interpreting context and relationships between concepts. Ontologies can be used to augment that understanding. If, for instance, an LLM generates an answer on a medical topic, medical ontology can contribute precise definitions of the concepts and the relationships between symptoms and diseases, allowing the LLM to understand better and generate more accurate responses.

LLMs are good at generating text, but they have no built-in knowledge structure. Ontologies provide that structure, making the more orderly organisation and management of knowledge possible. The data used by an LLM can be structured better and that, in turn, can give rise to results that are both more cohesive and more accurate.

LLMs also have none of the built-in, advanced inference mechanisms which are vital to numerous domains such as legal analysis or medical diagnosis. Ontologies make it possible to define the rules and relationships that LLMs can use for more complex inference. An LLM can use ontologies in a medical application, for example, in order to identify the relationship between symptoms and possible diagnoses and then generate appropriate recommendations.

Ontologies can be used to understand users’ individual preferences more precisely, allowing the relevant LLM to offer content which is more personalised. A recommendation system can serve as an example here. An ontology can assist in understanding which product categories are related and this then allows the LLM to suggest products that match up better with a given user’s preferences.

Neural networks are the foundations of numerous state-of-the-art AI applications, including image recognition, processing natural language and data prediction. Despite being powerful pattern recognition tools, they have their limitations. However, those limitations can be augmented by ontologies.

Neural networks learn from patterns in data, but they lack any kind of built-in domain knowledge. Ontologies can supply that domain knowledge, giving the networks an understanding of data context and significance. This connection can lead to results with greater accuracy.

Neural networks are frequently criticised for a lack of transparency in their decision-making processes. Ontologies can bring elements of symbolic inference to processes grounded in neural networks, facilitating decisions that are more comprehensible and explainable. Symbolic inference can work in tandem with deep learning, contributing rules and relationships that can be used by the networks for more complicated analyses.

In traditional neural networks, knowledge transfer between different tasks is limited. Ontologies can facilitate that transfer, enabling the networks to use knowledge from one domain in another, related domain. For instance, it will be easy for a neural network that has learned to identify objects in one ontology, like the vehicles ontology, to adapt that capability to another ontology which shares similar features, such as the agricultural machinery ontology. 

Ontologies are a tool that enables companies to refine and elevate the way they manage, organise and leverage information. They are used in a wide range of sectors, from e-commerce to biotechnology, and are extensively applied in the development of smart systems, data analysis and knowledge management. In this section, we present an overview of the types of companies that can benefit from bringing ontologies on board.

Companies operating in fields like finance, e-commerce, health care, logistics and telecommunications generate vast quantities of data. Ontologies aid them in organising, categorising and understand all that data, opening the door to more effective searches, analyses and decision-making. In the finance sector, for instance, they can assist with classifying transactions, managing risk and detecting fraud.

E-commerce is a sector where ontologies can significantly boost the customer experience and operational effectiveness. They expedite an improved understanding of customers’ preferences and offer personalisation and they improve product search results.

In biotechnology and pharmaceutical companies, ontologies are used for organising knowledge about genes, proteins, medications and chemical interactions. As a result, research can be conducted more rapidly and with greater accuracy, reducing the time needed to develop new therapies. Companies involved in making and improving medications can use ontologies to gain a better understanding of clinical trials and to predict the side effects of their products.

Companies in the industrial sector can use ontologies for managing knowledge of their products, production processes and supply chains. As an example, they can help in acquiring a better understanding of the relationships between machinery and equipment parts, opening up the possibility of more effective maintenance planning and reducing downtime.

Companies engaged in software development and delivering IT services can use ontologies to organise knowledge about system architectures, integrating applications and coding standards. They also simplify the management of complex projects where multiple teams have to collaborate on various aspects of the work while maintaining consistency and technical compliance.

Ontologies have enormous potential in business sectors ranging from e-commerce to the pharmaceutical industry. Companies that decide to deploy them can counts on improved data organisation, greater process effectiveness and decision-making capabilities with enhanced accuracy.

Ontologies in IT are advanced models for representing knowledge. They organise and structure information in a way that can be understood by people and machines alike. Created as a response to the need for improved data management and a means of facilitating effective communication between IT systems, they have a wide range of uses in numerous fields, from artificial intelligence and machine learning to knowledge management and data integration.

In the sphere of artificial intelligence, ontologies play a crucial role, providing context that means the operations of algorithms, including neural networks, can be more accurate and more comprehensible. As far as large language models, or LLMs, are concerned, ontologies support better understanding and natural language processing.

In business, ontologies can bring benefits to companies in a range of sectors, simplifying data management, the personalisation of services and process automation. Deploying them in businesses involves creating precise knowledge models that integrate and unify data from a variety of sources, which opens up a road to better decision-making. 

MakoLab is an expert in the field of ontologies. We offer advanced services, supporting companies in utilising the potential of ontologies to the full in the context not only of AI, but also of other technological applications.

What is deep learning?

Deep learning is an advanced machine learning method that uses neural networks to analyse and interpret large datasets. It is particularly effective for tasks like image recognition, natural language processing and recommendation systems. Deep learning learns data automatically, creating the potential for models that are more complex and more accurate.

What are neural networks?

Neural networks are computer systems inspired by the activities of the human brain. They consist of layers of nodes (neurons) that process input data and pass it on, making the creation of complex patterns possible. The foundation of deep learning, neural networks are used for image recognition, language processing and other spheres of AI.

What are large language models (LLMs)?

Large language models like GPT-4, BERT and T5, for example, use advanced learning techniques to process and generate natural language. They are trained with vast textual datasets, which enables them to generate texts and comprehensible, contextually appropriate answers to questions.

What is Web Ontology Language (OWL)?

OWL is a language used for defining ontologies and publishing them on the World Wide Web (WWW). Ontologies designed using OWL make it possible to represent complex relationships between concepts and to support the exchange of information between systems. OWL is a W3C standard and is widely used in fields like the semantic web, knowledge management and AI.

What is the Resource Description Framework (RDF)?

The RDF is a W3C standard for representing data on the World Wide Web. It facilitates the modelling of information in the subject-predicate-object form, known as a ‘semantic triple’, an ‘RDF triple’ or a ‘triple’. The technological cornerstone of the Semantic Web, the RDF allows date from a range of sources to be combined and processed in a way which can be understood by machines.

What is the World Wide Web Consortium (W3C)?

W3C is an international organisation that establishes standards for the World Wide Web (WWW). It is responsible for the development and promotion of standards like HTML, CSS, RDF and OWL, providing interoperability, cohesiveness and consistency in Web technology. W3C’s goal is the development of the Internet in a way which is open and accessible to everyone.

1. Wikipedia, Ontology
2. WC3 website, W3C 
3. V. Jain, Dr. S. V. A. V. Prassad, Ontology Based Information Retrieval Model in Semantic Web: A Review
4. Oxford English Dictionary, Ontology
5. Oxford Reference, Ontology
6. Merriam Webster, Ontology
7. IBM, What is a knowledge graph?

Translated from the Polish by Caryl Swift