RDF identifies the resources with a Uniform Resource Identifiers (URI) as described in [], but with slight difference: RDF make use of what we call as qualified URIs, that is, URIs with the optional fragment identifier (a text added to the URI with "#" between them). For [], the fragment identifier will return "the property of a data resulting from the retrieval action"; however, RDF do considers every qualified URI (with or without the fragment identifier) as the full resource by itself. In rest of this tutorial, we will therefore use a word "resource" in the RDF point of view, unless and otherwise specified.

The base element of RDF model is "triple" : a resource (subject) is linked to another resource (object) through arc labeled with third resource (predicate). We will say that the <subject> has the property <predicate> valued by <object>. Cosider for example, the triple in the figure below could be read as "Champion is the creator of an index.html".

All the triples result in the direct graph, whose nodes and arcs are labeled with the qualified URIs. Note in the figure below that the resource may have more than one value for a given property.

In RDF recommendation, targets of a graph can be pieces of text instead of the resources; these pieces of text are called the literals. Though this is syntactically useful in the RDF documents (being able to write some text directly rather than storing it in some other resource), handling it as an exception to the model is not appropriate, and that there is cleaner way of doing it: a piece of text probably has a fragment identifier so the literal node could be replaced by the standard URI node in the RDF graph. Note here that this alternative is compatible with recommendation as it uses the subset of a model.

RDF model is very simple and, above all, it is uniform. Mostly, the fact that the only vocabulary is URIs allows the use of same URI as a node and as arc label. This makes the things like self-reference and rectification possible, just like the natural languages. This is appreciable in user-oriented context (like web), but may become difficult to cope with the knowledge-based systems and the inference engines.