Inteligencia Artificial Semántica

Semantic technologies: the foundation of artificial cognition

Semantic technologies lie at the heart of our systems' ability to understand human reasoning. They form the foundation for the development of cognitive solutions that currently sit at the core of the artificial intelligence programme.

Within this paradigm, they enable systems to understand the way people reason and connect information, while simultaneously strengthening users' ability to comprehend texts, audio and discover relationships between data and entities. To achieve this depth of understanding, these technologies require a fundamental structure: knowledge graphs.

The way semantic technologies consolidate all that heterogeneous, distributed information in a meaningful way is by integrating it into a queryable knowledge graph modelled by an ontology. This graph relates entities to entities and entities to their attributes in a meaningful way, within the framework of a given knowledge domain.

But what exactly is an ontology? It is a formal description of knowledge that defines concepts within a specific domain — a bounded slice of reality — and the relationships between them. To construct this description, the ontology formally specifies components such as individuals (instances of objects), classes, attributes, relationships, constraints, rules and axioms. The result is a formal knowledge representation that is not only shareable and reusable, but can also derive new knowledge from what has already been represented. This derivation capability is made possible by the logico-formal languages used to declare these ontologies.

The standard language for formalising these ontologies is OWL (Web Ontology Language), a semantic markup language used to publish and share ontologies on the World Wide Web. It was developed by the World Wide Web Consortium (W3C) as an extension of RDF (Resource Description Framework) and RDFS (RDF Schema), providing richer semantics and greater knowledge representation capability than its predecessors.

OWL enables a knowledge domain to be rigorously modelled by defining classes, properties, relationships, constraints and axioms. Being grounded in first-order logic, OWL supports automatic inference and reasoning over the represented knowledge. This means that, in addition to modelling existing knowledge, OWL can infer new facts from existing ones — a capability that is fundamental to the development of intelligent applications.

One of OWL's key strengths lies in its expressiveness. The language provides a rich set of constructors for modelling complex knowledge, making it possible to define classes through necessary and sufficient conditions, specify cardinalities in relationships, define properties of relationships (transitivity, symmetry, etc.) and use logical connectors, among other things. To match this expressiveness to different computational needs, OWL is divided into three sub-languages, each with a different level of expressiveness: OWL Lite, OWL DL (Description Logic) and OWL Full.

Beyond its expressiveness, another fundamental characteristic of OWL is its ability to promote semantic interoperability. OWL ontologies can be published on the web, reused and linked to other ontologies, thereby fostering the exchange of knowledge between different systems and applications. This interconnection capability makes it a cornerstone of the semantic web and Linked Data vision.

This vision of interoperability is reinforced by the fact that, as a W3C standard, OWL integrates naturally with other semantic web technologies such as RDF, SPARQL and SKOS. OWL ontologies can be serialised in various formats such as RDF/XML, Turtle or Manchester Syntax, facilitating their adoption across diverse technical contexts.

The practical applications of OWL span a wide range of use cases, including the representation of taxonomies and controlled vocabularies, the modelling of domain-specific knowledge bases, semantics-based data and system integration, the development of semantic assistants and search engines, knowledge representation in expert systems, and interoperability across fields such as finance and banking, healthcare, bio and pharma, tourism and smart cities, and communications and media.

In short, OWL is a formal, standardised language designed to represent complex knowledge in a way that enables it to be processed and exchanged by computational systems on the web. Its ability to model knowledge domains in a rich and expressive way, combined with its logical foundations that support automatic reasoning, make it an essential tool for building intelligent systems based on knowledge representation and exploitation. Its adoption as a W3C standard and its integration with other semantic web technologies position it as a fundamental component in the infrastructure for knowledge management and exchange.

If OWL defines the logical structure of ontologies, RDF (Resource Description Framework) provides the fundamental model for representing the data that populates those ontologies. Developed and standardised by the W3C, RDF is the standard for describing web resources and exchanging data. Despite the existence of various conventional tools for managing data and their relationships, RDF stands out as the simplest, most powerful and most expressive standard designed to date.

RDF is the general method for describing information by defining relationships between data objects. Its power lies in its flexibility: it enables the effective integration of data from multiple sources, decoupling it from its original schema. This makes it possible to apply, interconnect, query and modify multiple schemas without changing the data instances. RDF is also built around established web standards: XML and URI.

RDF triple

The basic unit of RDF is the triple, also known as a "statement" or "RDF statement". An RDF triple is made up of three elements: subject, predicate and object. The subject represents the resource being described, the predicate defines the relationship between the subject and the object, and the object can be either another resource or a literal value.

This subject-predicate-object format can take any subject and connect it to any other object, using the predicate (verb) to indicate the type of relationship between them. For example, the statement "Las Meninas is in the Museo Nacional del Prado" can be represented as an RDF triple, where "Las Meninas" is the subject, "is in" is the predicate and "Museo Nacional del Prado" is the object. This model makes it possible to express virtually any fact or relationship through a uniform, linked structure.

This is how the RDF model increases the power of any piece of data by giving it the means to establish endless relationships with other pieces of data, becoming the basic building block of larger, more flexible and richly interconnected data structures. It is worth noting that all data, regardless of its original format, can be converted to RDF data. A central concept in this architecture is the Uniform Resource Identifier (URI), a unique identification system used on the web to identify resources unambiguously.

RDF for knowledge graphs

Knowledge graphs represented in RDF provide the best framework for integrating, unifying, linking and reusing data. This supremacy is grounded in five essential characteristics.

RDF graphs are used to manage Linked Open Data datasets such as DBPedia, which are published as RDF and interconnected. This enables federated queries to be run and answered more quickly, delivering highly relevant search results.

Beyond the representation and querying of explicit data, RDF graphs can infer implicit facts from explicit statements. With the help of a platform such as GNOSS Semantic AI Platform, this inference capability turns information into knowledge, enabling organisations to uncover hidden relationships in their data and close the loop connecting formal representation (OWL) with data instantiation (RDF) in a coherent knowledge management system.

Semantic interoperability: towards systems that truly understand one another

Semantic interoperability represents an advanced level of integration between systems that goes beyond mere technical connectivity. It refers to the ability of different systems to exchange data with a precise, shared meaning, ensuring that information is correctly interpreted by all components of the digital ecosystem.

For over two decades, GNOSS has been developing solutions that prioritise semantic interoperability as a fundamental element. Our experience has shown us that it is not enough for systems to connect — it is essential that they genuinely understand one another.

Technical interoperability refers to the ability of a system to connect with other systems in a standardised and efficient way. These connections enable functions such as data migration, data exchange between platforms, federated search and the removal of obsolete resources. Semantic interoperability, however, goes a step further: it ensures that exchanged data retains its precise meaning in every context.

To achieve this level of interoperability, knowledge graphs must be implemented and deployed using open standards such as RDF, SPARQL and OWL, promoted by the W3C. Using these standards facilitates interoperability and data exchange between systems, unlike proprietary implementations that can create technological dependencies and information silos.

Semantic interoperability enables comprehensive data management, ensuring not only that data can "talk to" one another within a system, but also that they interact effectively with third-party systems. This capability enriches and contextualises information, improving its accuracy, usefulness and reach.

Ensuring that every system can share information fully, accurately and in a timely manner allows organisations to make the most of the capabilities their systems were designed for. Conversely, the absence of interoperability leads to systems that grow in isolation, creating a fragmented data ecosystem for both the organisation and its users. These interoperability functions, embedded in a Semantic Data Fabric solution, help organisations combat "data entropy" — the tendency of data ecosystems to become more chaotic and disorganised over time.

Semantic interoperability rests on two complementary sets of principles that lay the foundations for a truly connected data ecosystem: the FAIR principles and the Linked Open Data principles.

The FAIR principles establish that data must be:

• Findable: they must have unique and persistent identifiers (such as URIs) and be described with structured metadata that allows them to be located and interpreted.
• Accessible: they must be retrievable through universal protocols such as HTTP or HTTPS, allowing secure access via digital certificates, and guaranteeing that metadata remains available even if the original data no longer is.
• Interoperable: they must use standardised, formalised vocabularies — such as OWL and RDF — that are widely adopted and understood by different systems and communities.
• Reusable: they must be published under clear licences that define their conditions of use, include information about their provenance, and comply with the standards of the relevant domain, enabling their application in different contexts without loss of reliability.

The Linked Open Data principles, established by Tim Berners-Lee, the creator of the World Wide Web, constitute a set of rules that define the technologies and practices needed to publish data in a way that drives the creation of a globally interconnected space. These principles require:

• Using URIs (Uniform Resource Identifiers) as unique identifiers for entities.
• Using HTTP URIs to enable direct lookup and access to these identifiers.
• Providing useful information when a URI is queried, using standards such as RDF or SPARQL.
• Including links to other URIs, facilitating the connection and discovery of further related data.

Publishing data as Linked Data is an optimal way to put the FAIR principles into practice, as it directly addresses the key aspects of each principle. Linked data uses URIs as unique identifiers (first principle), and its association with the HTTP protocol (second principle) makes data findable on the web. The description of and relationship with other resources (third principle) is achieved through the application of the RDF model, while retrieval and manipulation take place via the SPARQL protocol. The fourth principle, together with the third, is responsible for the effective linking of data from different sources, reinforcing the idea that the benefits of linking data are amplified when that data is open.