Understanding Ontologies¶

An ontology is a formal representation of knowledge within a domain. It defines the types of entities that exist, their properties, and the relationships between them.

What is an Ontology?¶

In the context of knowledge graphs, an ontology serves as a schema or blueprint that describes:

• Classes: The types of entities (e.g., Person, Organization, Aircraft)
• Properties: Attributes of entities (e.g., name, date, location)
• Relationships: How entities connect to each other (e.g., WORKS_FOR, LOCATED_IN)
• Constraints: Rules about valid combinations (e.g., a Person can have only one birth date)

Ontology (Schema)              Knowledge Graph (Data)
─────────────────              ─────────────────────
Person                    →    John Smith
  - name: string               - name: "John Smith"
  - birthDate: date            - birthDate: 1985-03-15

Organization              →    Acme Corp
  - name: string               - name: "Acme Corp"
  - founded: date              - founded: 2010-01-01

WORKS_FOR                 →    John Smith ──WORKS_FOR──► Acme Corp
  - since: date                - since: 2020-06-01

Why Ontologies Matter¶

1. Consistency¶

Without an ontology, the same concept might be extracted differently:

An ontology ensures that semantically equivalent concepts map to the same canonical type.

2. Interoperability¶

Ontologies enable data from different sources to be integrated:

Source A: "Boeing 737-800"     ─┐
Source B: "B738"               ─┼──► Aircraft (type: Boeing 737-800)
Source C: "737-800 aircraft"   ─┘

When multiple systems use the same ontology, their data becomes automatically compatible.

3. Domain Expertise Capture¶

Ontologies encode expert knowledge about a domain:

• Aviation: An "Occurrence" involves Aircraft, happens at an Airport, has a Flight Phase
• Financial Crime: A "Sanction" targets an Entity, issued by an Authority, has a Program ID
• Healthcare: A "Diagnosis" relates to a Patient, made by a Physician, has an ICD code

This expertise guides extraction and ensures domain-relevant relationships are captured.

4. Query Intelligence¶

With an ontology, systems can understand that:

• A query for "airlines" should match entities of type Operator
• A query for "plane crashes" relates to Occurrence with certain Events
• A query for "sanctioned companies" means Organization with SANCTION relationship

Ontology vs Schema vs Taxonomy¶

An ontology is the most expressive, supporting:

• Inheritance: A CommercialAircraft inherits properties from Aircraft
• Inference: If A is part of B, and B is located in C, then A is located in C
• Constraints: An Occurrence must have exactly one primary_cause

Standard Ontology Formats¶

OWL (Web Ontology Language)¶

The W3C standard for ontologies, typically serialized as Turtle (.ttl):

@prefix ex: <http://example.org/> .
@prefix owl: <http://www.w3.org/2002/07/owl#> .
@prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .

ex:Aircraft a owl:Class ;
    rdfs:label "Aircraft" ;
    rdfs:comment "A vehicle capable of flight" .

ex:Operator a owl:Class ;
    rdfs:label "Operator" ;
    rdfs:comment "An organization that operates aircraft" .

ex:operatedBy a owl:ObjectProperty ;
    rdfs:domain ex:Aircraft ;
    rdfs:range ex:Operator .

FollowTheMoney (FTM)¶

A schema designed for investigative journalism and anti-corruption:

Person:
  properties:
    - name
    - birthDate
    - nationality

Organization:
  properties:
    - name
    - jurisdiction
    - registrationNumber

Ownership:
  properties:
    - owner: Person | Organization
    - asset: Organization
    - percentage

Domain-Specific Ontologies¶

Different domains have established ontologies:

Using established ontologies provides:

• Standardization: Industry-accepted terminology
• Completeness: Years of domain expert refinement
• Interoperability: Data exchange with other systems

Aletheia's Ontology Processing¶

Aletheia's GenericOntologyLoader processes ontologies with several capabilities beyond basic OWL parsing:

• Transitive class classification — Determines whether each class is an entity type, relationship type, or abstract class by checking the full ancestry chain (not just direct parents).
• Non-reified relationship extraction — Object properties between entity classes are extracted as relationship types alongside reified relationship classes.
• ModelingProfile support — Optional explicit classification hints that override heuristic rules, useful when an ontology's structure doesn't fit the default patterns.
• Enriched docstrings — Property names and descriptions from the ontology are appended to entity type docstrings, giving the LLM concrete extraction signal.

See Integration for the full workflow.

Next Steps¶

• Business Value of Ontologies — ROI and practical benefits
• Aletheia Integration — How Aletheia uses ontologies for extraction
• Use case specific ontologies:
  • ECCAIRS Aviation Taxonomy
  • FollowTheMoney Schema