HarryPotter hasPet Hedwig . variables in first-order logic.
where HarryPotter is the subject, A set of triples is called an RDF graph
hasPet is the predicate, and Hed- (see Figure 1). In order to facilitate the
wig is the object. The subject of a sharing and exchanging of graphs on
triple is either an IRI or a blank node the Web, the RDF specification in-
(an unlabeled node), while the ob- cludes an XML serialization. In RDF/
ject is an IRI, a blank node, or a lit- XML thetriplescanbe writtenas
eral value (such as a string or integer). <rdf:Description
For example, we could use the triple: rdf:about=”#HarryPotter”>
HarryPotter hasemail <hasPet
“ firstname.lastname@example.org”. rdf:resource=”#Hedwig”/>
to capture information about Harry’s <hasEmail>harry.potter@
email address. The predicate of a triple
is always an IRI called a “property.” IRIs </hasEmail>
are treated as names that identify par- </rdf:Description>
ticular resources. Blank nodes also de- where #HarryPotter and #Hedwig
note resources, but the exact resource are fragment identifiers.
being identified is not specified, behav- The RDF specification also extends
ing instead like existentially quantified the capabilities of the language by giv-
ing additional meaning to certain re-
sources. One of the most important is
rdf:type, a special property that cap-
tures the class-instance relationship;
where rdf is an abbreviation (called
a “namespace prefix”) for the string
For example, we could use the triple:
immaterial HarryPotter rdf:type Wizard .
to represent the fact that Harry is an in-
stance of Wizard.
RDF provides a flexible mechanism
Spirit for adding structured annotations but
does little to address the problem of
understanding the meaning, or se-
mantics, of the terms in annotations.
One possible solution would be to fix
a set of terms to be used in annota-
tions and agree on their meaning. This
works well in constrained settings like
annotating documents; the Dublin
Core Metadata Initiative (dublincore.
org/schemas/) defines just such a set
of terms, including, for example, the
properties dc:title, dc:creator,
dc:subject, and dc:publisher.
However, this approach is limited with
respect to flexibility and extensibility;
only a fixed number of terms is defined,
and extending the set typically requires
a lengthy process in order to agree on
which terms to introduce, as well as on
their intended semantics. It may also
be impractical to impose a single set of
terms on all information providers.
An alternative approach is to agree
on a language that can be used to de-
fine the meaning of new terms (such
as by combining and/or restricting ex-
isting ones). Such a language should
preferably be relatively simple and pre-