Standards

Resource Description Framework (RDF)

A graph data model for statements about identified resources.

Kind: Data models
Steward: World Wide Web Consortium (W3C)
Official source: https://www.w3.org/TR/rdf11-concepts/

What it describes

RDF represents information as subject-predicate-object statements. IRIs identify resources and properties, while literals carry values such as strings and dates. The model supports combining statements from different publishers when their identifiers and meanings are aligned. It is the basis for web vocabularies such as SKOS and DCAT, and it has several concrete serialization formats.

Sharing RDF syntax does not make two predicates equivalent or two similarly named resources identical. RDF also permits incomplete descriptions; absence of a triple is not generally evidence that its proposition is false. For evidence-sensitive use, preserve named source context, vocabulary definitions, and timestamps. Treat identity links and inferred statements as claims with provenance rather than collapsing them into unexplained facts.

Structure

Graph of triples with IRI, blank-node, or literal terms under RDF specifications.

Useful for

Combine graph-shaped public data.
Express typed relationships.
Use shared vocabularies across publishers.

Limits

Shared syntax does not ensure shared meaning.
Missing triple is not a negative fact.
Identity links require review.

How it is used in practice

RDF turns every piece of information into a statement of three parts: a subject, a predicate and an object. Subjects and predicates are named by IRIs, objects are IRIs or literals with a datatype or language tag, and blank nodes stand for things that have no global name. Collections of statements form a graph, and collections of graphs form a dataset in which each graph can have a name. The model has several serializations, including Turtle, N-Triples and JSON-LD, which can express the same graph. Vocabularies such as RDF Schema and OWL give predicates and classes defined meaning, and a query language, SPARQL, retrieves and combines statements.

Because identifiers are global, graphs published separately can be merged without renaming. This is the main attraction for knowledge fusion: statements from different sources about the same IRI fall together, and statements about different IRIs stay apart until a link says otherwise.

A worked example

Two fictional sources publish statements about Halden Therapeutics AG. The register says the company has a headquarters in one city. A news dataset says the company has a headquarters in another. Both use the same IRI for the company, so a merge puts both statements on one node.

A query for the headquarters returns two values with no explanation. If each source was loaded into its own named graph with metadata about the source and the retrieval date, the query can return the graph name too. The reviewer then sees which source said what and when. The model does not decide which is right. It allows the disagreement to be represented, and the decision to be made by an explicit rule that is itself recorded. Named graphs and similar devices are what turn a pile of triples into evidence.

Evaluating a claim that rests on RDF data

Ask where the statements came from. A triple has no author, date or confidence built in, so any of these must be attached through a graph, a reification pattern or a convention. A graph that arrives without them is a set of unsourced assertions. Ask what the IRIs mean. Two predicates that look alike can differ in meaning, and a predicate's definition lives in its vocabulary, which may have changed since the data was published.

Ask what the absence of a triple means. RDF assumes an open world: a statement not present is not thereby false. A query that treats absence as negation is applying a closed-world reading that the data does not license. If a negative is needed, it has to be asserted, with its basis.

Confusions to avoid

RDF is a data model, not a file format and not a database product. The same graph can be stored in many ways, and the choice of serialization or store affects performance and tooling but not meaning. It is also not a guarantee of interoperability. Using RDF does not make two datasets compatible. Alignment of their vocabularies and identifiers is separate work.

A final confusion is between identity links and facts. A statement that two IRIs denote the same thing is a strong claim, and merging on the strength of such links can contaminate everything attached to either node. Treat identity links as claims with sources and review them like any other consequential statement. A newer revision of the model, RDF 1.2, adds triple terms to make statements about statements easier and is at an advanced stage at W3C.

Questions a reviewer can ask

Which graph holds each statement, and what metadata does the graph carry? What do the predicates mean in their current vocabulary version? Are identity links asserted, by whom, and on what evidence? Does any query treat missing statements as negation? Was the graph retained as received? Can the answer to a query be explained by listing the statements and sources it used?

Related terms

References

Found an error on this page? See corrections for what counts as one, how to report it and the dated log.

Enquiry

Which of your sources carry Resource Description Framework (RDF)?

Tell us which sources carry it, which key the same entity some other way and who may read which. We reply with whether and how a first record would use this data model.

Open a structured enquiry Contact and what happens next

Or write to [email protected] in your own words.