The Layered Approach to Ontology Engineering: Building Knowledge Systems That Scale

In today's data-driven world, organizations are increasingly turning to ontologies to bring structure and meaning to their information assets. As someone who has worked with knowledge graphs and semantic technologies for over a decade, I've seen firsthand how a well-designed ontology can transform data into actionable insights.

Today, I want to share a powerful framework for ontology engineering that addresses one of the biggest challenges in the field: balancing generalizability with specificity.

The Ontology Pyramid: A Strategic Framework

The ontology pyramid, adapted from the work of Armin Haller and Axel Polleres (2020), provides a comprehensive approach to building knowledge systems that are both robust and flexible. Let me walk you through each layer and explain why this structured approach matters for your organization.

Upper Ontology

At the top of our pyramid sits the Upper Ontology - the most abstract and context-independent layer. This foundation defines universal concepts like time, space, and events that transcend specific domains. Think of upper ontologies like DOLCE, BFO, or SUMO that provide the philosophical underpinnings for your entire knowledge architecture.

Key characteristic: Low number of concepts but high context independence and applicability across domains.

Domain Ontology

Moving down, we encounter Domain Ontologies, which come in two varieties:

1. Upper Domain Ontology: These capture general concepts within a specific field (like healthcare, finance, or manufacturing).
2. Core Domain Ontology: These define the essential concepts and relationships that are central to your domain.

Domain ontologies strike a balance between generalizability and specificity, providing the conceptual backbone for your knowledge system.

Subdomain Ontologies

This layer gets more specialized, focusing on particular aspects within your domain. The diagram shows Complex Subdomain 1, Complex Subdomain 2, and Other Subdomains - each with its own network of interconnected concepts. These subdomain ontologies allow for detailed modeling of specific areas while maintaining connections to the broader domain concepts.

Application Ontologies

At the bottom of the hierarchy (but not the bottom of importance!) are Application Ontologies. These are highly specialized and context-dependent, designed to support specific applications, products, or use cases. They connect directly to the data products your organization creates and manages.

Why This Layered Approach Matters

This hierarchical approach to ontology engineering delivers several critical benefits:

1. Scalability: As your data ecosystem grows, the layered structure allows you to extend your ontologies without breaking existing systems.
2. Reusability: Upper and domain ontologies can be reused across multiple projects, saving tremendous development time.
3. Federated Modeling: Different teams can work on different sections of the ontology without stepping on each other's toes.
4. Maintenance Efficiency: Changes can be isolated to specific layers, minimizing ripple effects throughout your knowledge base.
5. Knowledge Integration: The connections between layers (represented by the arrows in the diagram) facilitate seamless integration of knowledge across your organization.

Putting This Into Practice

In my experience implementing this approach with clients across industries, I've found that successful ontology programs follow these principles:

• Start broad, then narrow: Begin with upper and domain ontologies before diving into specifics.
• Connect the dots: Ensure clear linkages between layers through well-defined relationships.
• Involve domain experts: Subject matter experts are essential, especially at the subdomain and application levels.
• Think about governance: Establish clear ownership for each layer of the ontology pyramid.

Ontologies as the Semantic Foundation for Data Mesh

As @andrea giola insightfully points out:

"Data only realizes its value when it is used—the more it is used, the more valuable it becomes. Its value grows even further when it is combined with other data and then utilized."

This observation strikes at the heart of why ontologies matter in modern data architectures, particularly in the Data Mesh approach that many organizations are adopting today.

One of the biggest challenges in the Data Mesh paradigm isn't just creating data products but ensuring they can be reused and composed across domains. The main obstacle, as Gioia notes, is the "semantic gap" that traps data products in different semantic silos.

The layered ontology approach addresses this challenge directly:

• Upper and Domain ontologies serve as the shared semantic foundation that allows data products from different domains to "speak the same language"
• Subdomain ontologies enable specialized teams to define concepts relevant to their area while maintaining connections to the broader semantic framework
• Application ontologies connect directly to data products, providing the semantic context needed for meaningful integration

While building a shared enterprise ontology can indeed be organizationally complex, the pyramid structure offers a pathway that's both scalable and practical. It allows for federated modeling teams (as shown in the diagram) to work independently on their domains while ensuring semantic coherence across the enterprise.

Looking Ahead

As data volumes continue to explode and AI systems become more integrated into business processes, well-structured ontologies will only grow in importance. Organizations that invest in thoughtful ontology engineering today will be better positioned to extract value from their data assets tomorrow.

The beauty of the ontology pyramid is that it provides both stability through its upper layers and flexibility through its lower layers - exactly what organizations need in today's rapidly evolving digital landscape, particularly as they embrace data-driven approaches like Data Mesh.

What has been your experience with ontology engineering? Have you implemented a layered approach like this in your organization? I'd love to hear your thoughts and experiences in the comments.

This article draws inspiration from the ontology hierarchy model developed by @Armin Haller and Axel Polleres (2020), which provides a structured approach to balancing abstraction and specificity in knowledge representation. Special thanks to Andrea Gioia for the valuable insights on data mesh and semantic integration that helped shape this discussion.

-Written by The Innovater