SPARQL Distinct pairs - sparql

Question

I've got a table, where there are documents with identical authors. I need to get the distinct pairs of documents. I did the following:
SELECT DISTINCT ?d1 ?d2 WHERE {
?d1 myns:creator ?x.
?d2 myns:creator ?y.
FILTER (?x=?y && ?d1!=?d2).
}
GROUP BY ?d1 ?d2
But for this both DOC1, DOC2 and DOC2, DOC1 are in the result. I need to get rid of one of the pairs.
Here is the whole triples database:
#prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
#prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
#prefix myns: <http://my.local.namespace#> .
_:doc1 rdf:type myns:Document.
_:doc1 myns:creator _:Pete.
_:doc1 myns:year "2000"^^xsd:integer.
_:doc1 myns:publisher _:p1.
_:doc2 rdf:type myns:Document.
_:doc2 myns:creator _:John.
_:doc2 myns:year "2004"^^xsd:integer.
_:doc2 myns:publisher _:p2.
_:doc3 rdf:type myns:Document.
_:doc3 myns:creator _:Pete.
_:doc3 myns:publisher _:p3.
_:doc4 rdf:type myns:Document.
_:doc4 myns:creator _:Bob.
_:doc4 myns:year "2010"^^xsd:integer.
_:doc4 myns:publisher _:p2.
_:Pete rdf:type myns:Person.
_:Pete myns:knows _:Bob.
_:Pete myns:knows _:John .
_:John rdf:type myns:Person.
_:John myns:age "29"^^xsd:integer.
_:John myns:knows _:Bob.
_:Bob rdf:type myns:Person.
_:Bob myns:age "35"^^xsd:integer.
The result, that I am getting, after executing query is:
D1 D2
_:891f1e98-b411-4e54-9533-18d530f09c6ddoc1 _:891f1e98-b411-4e54-9533-18d530f09c6ddoc3
_:891f1e98-b411-4e54-9533-18d530f09c6ddoc3 _:891f1e98-b411-4e54-9533-18d530f09c6ddoc1
As it is noticeable, technically both pairs are same. I junst need distinct one (i.e. one of them is enough). I am not sure about enviromental characteristics. But there is Sesame framework

Answer 1

This will work in some systems:
SELECT ?d1 ?d2 WHERE {
?d1 myns:creator ?x.
?d2 myns:creator ?y.
FILTER (?x=?y && STR(IRI(?d1)) < STR(IRI(?d2))).
}
?d1 and ?d2 are going to be blank nodes. But blank nodes are blank.
So to provide the ordering for <, we need some kind of query-wide label or value associated with each one.
Your data does not have any distinguishing triples for each person.It would be better to put real names in the data:
_:Pete rdfs:label "Pete" .
Even better, use the FOAF vocabulary.
Some systems allow blank nodes in IRI() - technically it's an extension of the SPARQL specification. You can then take the STR form and compare. that works on your data for me (Apache Jena) - You don't say which RDF system you are using.
The best solution is put distinguishing information into the data.

Answer 2

You can do this with a little trick: turn the != into a < (or >), and convert the values to strings, so you can do lexical comparisons:
SELECT DISTINCT ?d1 ?d2 WHERE {
?d1 myns:creator ?x.
?d2 myns:creator ?y.
FILTER (?x=?y && STR(?d1) < STR(?d2)).
}
GROUP BY ?d1 ?d2
This works on the idea that for any pair of identifiers that are not equal, one identifier is always greater than the other (by lexical ordering). So of any two pairs, only one will actually be selected.
Update so now that you've shown your data, we can see that the problem is that you are not using IRIs to distinguish your documents, but use blank nodes. The above query does not work because according to the SPARQL standard, blank nodes are unordered (so directly comparing via < does not work), and moreover, the STR function is defined to only operate on literals or IRIs, not on blank nodes.
The best solution is to change your data, and make sure that you use proper IRIs, because regardless of whether you can somehow make this query work on this data, the result would be almost useless: blank nodes have no meaning outside their local scope, so the document identifiers that your query returns can not really be reused; for example, you would not be able to do a SPARQL query that gets any properties specifically for the document _:doc1 (although to be fair Sesame has a workaround for this in the API).
A very simple way to change your blank nodes to IRIs, by the way, is to replace all occurrences of _: in your turtle file with myns:.