Best Practice: Writing SPARQL Queries that Compensate for Alternate Types of String Literals

State: Approved

Contact: Nick Crossley

Scope

In RDF there are three alternative ways that a literal value can be encoded. A literal may be either plain or typed. Furthermore, a plain literal may have an optional language tag. These alternative values are distinct. Furthermore, for typed or tagged values to be equal, their types or tags must also be equal. These differences might result in unexpected query results unless you take care in writing the query.

Recommendation

The solution is to use the SPARQL STR() function to extract just the string part of the value. SPARQL has other functions for extracting the type and language tag, e.g. if you just want English values.

Examples

To illustrate this, consider the following example resource <http://example.com/ex1/> where the string "title" is represented three ways:

@prefix dcterms: <http://purl.org/dc/terms/> .
@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .

@base <http://example.com/ex1/> .

<plain> dcterms:title "title" .

<typed> dcterms:title "title"^^xsd:string .

<english> dcterms:title "title"@en .

This example contains 3 triples, as shown by this query:

select *
where  {
   graph <http://example.com/ex1/> {
      ?s ?p ?o .
   }
}

Now suppose we write a query to find all the resources whose dcterms:title is "title":

select *
where  {
   graph <http://example.com/ex1/> {
      ?s ?p "title" .
   }
}

The result is just the triple with the plain literal value.

We can use a FILTER clause to get more triples:

select *
where  {
   graph <http://example.com/ex1/> {
      ?s ?p ?o .
      filter(?o = "title") .
   }
}

The definition of the SPARQL equals (=) operator does some conversion, but omits the tagged value. To get all values, use the SPARQL STR function.

select *
where  {
   graph <http://example.com/ex1/> {
      ?s ?p ?o .
      filter(str(?o) = "title") .
   }
}