Knowledge Graph Schema Generation from Text

Automatically derive structured graph schemas from natural language documentation and domain descriptions.

This skill converts textual descriptions into machine-readable graph models with entities, relationships, properties, and constraints.

Quick Start

Use When

• Converting domain documentation → ontology
• Bootstrapping a knowledge graph schema
• Designing initial RDF/OWL or Neo4j schemas
• Extracting schema from requirements or API docs

Inputs

• Natural language domain descriptions
• Technical documentation
• Example records or datasets
• JSON/CSV structures

Outputs

• Entity types (nodes/classes)
• Relationship types (edges/predicates)
• Properties & attributes
• Graph schema representation (Property Graph or RDF)

Example

Input:

A university contains students, professors, courses, and departments.
Students enroll in courses. Professors teach courses.
Departments manage both courses and professors.

Output:

Entities: Student, Professor, Course, Department

Relationships:
- Student -> ENROLLED_IN -> Course
- Professor -> TEACHES -> Course
- Department -> MANAGES -> Course
- Department -> MANAGES -> Professor

Properties:
- Student: id, name, email, enrollment_date
- Course: id, title, credits, department
- Professor: id, name, department, specialization
- Department: id, name, budget

Execution Steps

1. Extract Entities – Identify nouns/concepts from text
2. Extract Relationships – Identify verbs/connections
3. Extract Properties – Identify attributes and constraints
4. Infer Structure – Build graph patterns
5. Generate Schema – Output in target format (Property Graph or RDF)

Schema Formats

Property Graph (Neo4j, TigerGraph)

Nodes: Student, Professor, Course, Department
Relationships: ENROLLED_IN, TEACHES, MANAGES
Properties: Names, IDs, dates, descriptions

RDF/OWL (Semantic Web)

Classes: Student, Professor, Course, Department
Properties: enrolledIn, teaches, manages
Attributes: name, id, description

Recommended Libraries

• NLP: spaCy, transformers, nltk
• Graph: networkx, rdflib, pyvis, owlready2
• Schema: pydantic, dataclasses, jsonschema

Best Practices

✓ Use clear, consistent entity names (PascalCase)
✓ Normalize relationship directions
✓ Extract domain-specific constraints
✓ Separate schema from instance data
✓ Follow knowledge graph modeling standards

References

See extraction-patterns.md for entity/relationship extraction guidelines and example-schemas.md for domain examples.

Version: 1.0.0