The semantic web and ontologies

The web was built for people: a page links to another page, but the link means nothing to a machine beyond "follow this". The semantic web idea is to publish data in a form that carries meaning — not just documents that mention courses and students, but statements a program can combine and reason with.

The unit of that data is the triple. Everything else — ontologies, query languages, reasoning — is built on top of it.

The widget below holds a small university ontology, toggles RDFS inference on and off, and lets you query it with a triple pattern.

Triples, IRIs and vocabularies #

A triple (subject, predicate, object) uses globally unique identifiers (IRIs) so that terms from different publishers mean the same thing. A small set of standard vocabularies does most of the work:

• RDF gives the model: subject predicate object, plus rdf:type.
• RDFS adds schema vocabulary: rdfs:Class, rdfs:subClassOf, rdfs:domain, rdfs:range, rdfs:label.
• OWL adds expressive logic: owl:equivalentClass, owl:disjointWith, cardinality restrictions, property characteristics (transitive, symmetric, functional).

An ontology is a set of schema triples describing a domain; the knowledge graph is the instance data plus that schema.

Inference: deriving what you did not store #

RDFS entailment is a small rule set. Three rules do most of the visible work:

1. Subclass transitivity: from (A subClassOf B) and (B subClassOf C), infer (A subClassOf C).
2. Domain: from (p domain C) and (x p y), infer (x type C).
3. Range: from (p range C) and (x p y), infer (y type C).

Plus type propagation: from (x type C) and (C subClassOf D), infer (x type D). Repeat to a fixed point and new triples appear that no one wrote down. OWL extends this with decidability-aware profiles (EL, QL, RL) tractable enough for real knowledge graphs.

SPARQL #

Queries are graph patterns with variables. A pattern like

SELECT ?course WHERE {
  ?student :enrolledIn ?course .
  ?course rdf:type :Course .
}

matches every way of binding ?student and ?course in the store. Pattern matching over triples is the query engine; the reasoning above happens either eagerly (materialise inferred triples) or lazily during query rewriting.

Why it did not take over the web #

The original vision assumed publishers would annotate everything. Incentives were weak, vocabularies multiplied, and the open-world assumption — absence of a triple is not falsity — confuses people expecting database semantics. What survived is the powerful part: knowledge graphs inside search engines, recommenders and enterprise systems, almost always with closed-world validation (SHACL) and entity linking rather than open-web annotation.

Where next: from networks of people and facts to networks of pixels and physiology — medical imaging and wearables.