Semantic Web Academy — home

Instance data vs ontology layer

Two layers of triples — and why mixing them up costs you later.

0/3 done

Theory

A mature KG has two layers of triples that live in the same graph but answer different questions:

LayerExample tripleQuestion it answers
Instance:Alice :knows :BobWhat is true right now?
Ontology:knows rdfs:domain :PersonWhat does this predicate mean?
Ontology:Person rdfs:subClassOf :AgentHow are concepts related?

Keep them in separate named graphs so you can: (a) ship the ontology with the codebase but let instance data churn; (b) reason over the ontology to infer new instance triples; (c) version them independently.

Visualization

Two layers in one KG: instances on the bottom, ontology on the top. An LLM that retrieves both layers can answer questions the instance data alone can't (e.g. is Alice an Agent? — yes, via the subClassOf chain).

Reading in progress · 0 of 3 activities done