What Are Ontology-based Patent Classifiers and How to Use Them in Your Company

In the landscape of Patent Intelligence for tech companies, properly mapping and organizing technical information has become a high-value strategic factor. However, the growth in document volumes and the complexity of technical language increasingly requires the support of advanced technologies to accurately classify this data. Natural Language Processing applied to patents comes to our aid.

In recent years, AI for IP has shown significant potential in the field of patent classification. It is possible to accelerate the automatic assignment of patents to categories of interest, but as highlighted in previous articles, general-purpose models do not always fully capture the technical-engineering meaning of inventions. This can lead to classification errors or the loss of critical information, a risk that must be carefully managed given the potential legal and economic consequences of incorrect categorizations.

It is precisely to bridge this gap that Ontology-based Patent Classifiers have emerged: tools designed to understand technology not as generic text, but as a system governed by functions, relationships, and engineering principles, enabling effective mapping and precise categorization. Let’s take a deeper look at these tools for patent classification in a corporate context.

How Does a Technical Ontology Work in Patent Classification?

An ontology is a formal knowledge model that describes concepts, properties, rules, and constraints. The difference between an ontology and a taxonomy is that the latter organizes concepts into a hierarchy, while an ontology also represents the connections between them, offering a much richer and more detailed view of the domain.

An ontology-based classifier therefore uses a structured representation of technical knowledge, built not through statistical correlations but through conceptual models consistent with engineering logic.

In this context, an ontology is a formal map of knowledge within a technological domain: it describes a technology’s functions, components, causal relationships between parts, problems it addresses, and the physical principles governing its operation. Each technological sector has specific entities and relationships, and therefore requires a dedicated ontology.

When applied to patent classification, the presence of a domain ontology allows the system to interpret inventions in context for what they actually are, not merely based on the terms used to describe them, overcoming the limitations of keyword-based classification. This means the classifier can recognize a technical concept even if expressed through linguistic variations or complex descriptions, because it identifies it based on its function and purpose.

Why Are Ontology-based Patent Classifiers Different from General-purpose LLMs?

A general-purpose LLM does not truly understand the meaning of texts: it predicts which word is most likely to follow another based on patterns observed in millions of examples, producing plausible responses. Engineering, however, is not about probability but rigor: it is based on functions, constraints, and cause-effect relationships. By integrating technical ontologies into LLM training, the model becomes truly effective in engineering contexts, increasing task accuracy from the typical 50% of general models to up to 95%.

It is precisely on this premise that Erre Quadro’s ontology-based classifiers were developed, where patent classifications no longer depend on how a technology is mentioned in the text, but on the functions it performs.

This approach to LLM training allows companies to fully harness the power of AI, significantly accelerating activities without sacrificing the necessary rigor. It eliminates many of the typical pitfalls of general models, such as information loss, failure to detect weak signals, or the generation of inaccurate data, providing a complete, precise, and reliable representation of the technological domain.

How Can a Company Integrate Ontology-based Patent Classifiers into Its Processes?

The integration of an Ontology-based Patent Classifier must be tailored to the internal needs of the company, as each organization has different processes and strategic priorities.

In the implementation of Innovation Reveal®, for example, classification can follow two distinct approaches. The first is a top-down approach, in which the classification structure mirrors the “categories” already used by the client: product lines, strategic technology areas, application domains, or R&D processes. This ensures immediate alignment with the company’s language and decision-making framework, promoting rapid and consistent use of results. The second is a bottom-up approach, starting directly from patent data and observing categories that emerge spontaneously through the clustering of functions, solutions, and technical problems addressed by inventions. This method makes the invisible visible: uncovering hidden patterns, uncharted technological areas, weak signals, and new directions for development that were previously overlooked… in other words, revealing hidden innovation.

What Is the Strategic Value of Applying Ontology-based Patent Classifiers?

Adopting an Ontology-based Patent Classifier allows companies to radically transform the way they manage and interpret patent information, optimizing patent analysis in R&D. It introduces a classification system capable of reliably grouping all truly relevant technical documents, accelerating searches for patents on specific, even complex, focuses, facilitating internal sharing, and supporting faster, more informed decision-making.

The resulting value is significant. Companies now hold a map of their technological domain: they are no longer interpreting hundreds or thousands of isolated documents, but navigating a conceptual structure that fully represents the innovative landscape of interest. This ability to turn data into actionable knowledge enables companies to move from a reactive to a truly proactive view of innovation.