RDF Triple Stores

RDF (Resource Description Framework) represents everything as triples:

(subject, predicate, object)

(:alice, :name, "Alice")
(:alice, :knows, :bob)
(:alice, :worksAt, :acme)

URIs identify resources; literals are strings/numbers/booleans.

Knowledge graphs: hundreds of billions of triples.

Query language: SPARQL:

SELECT ?friend
WHERE {
  :alice :knows ?friend .
  ?friend :age ?age .
  FILTER (?age > 25)
}

Compared to property graph:

• Triples are atomic; no edge properties (need reification).
• Standardized via W3C.
• Used by linked data (Wikidata, DBpedia, science ontologies).
• More verbose for app development; better for interop + semantics.

Storage:

• Triple table: (s, p, o, context).
• Indexes: SPO, POS, OSP at minimum (different access patterns).
• Compression: dictionary-encode URIs (most repeat).

Reification (adding metadata to triples):

:alice :knows :bob
+
:s1 :rdfType :Statement
:s1 :subject :alice
:s1 :predicate :knows
:s1 :object :bob
:s1 :since 2010

Verbose. RDF-star (newer extension) makes this less painful.

Inferencing:

• Class hierarchies (Person is a Mammal).
• Transitive properties (A knows B, B knows C → A knows C? maybe not).
• OWL reasoners do entailment.

Property graph: app-developer friendly, fast, ad-hoc schemas. RDF: standardized, semantic, interoperable, slower.

Most modern apps pick property graph. Knowledge engineering tools pick RDF.