1

I have an instance of Wikibase running. I'm able to run federated queries with Wikidata successfully. I have certain queries that compare labels like this:

PREFIX xwdt: <http://www.wikidata.org/prop/direct/>
PREFIX xwd: <http://www.wikidata.org/entity/>
PREFIX xpq: <http://www.wikidata.org/prop/qualifier/>
PREFIX xps: <http://www.wikidata.org/prop/statement/>
PREFIX xp: <http://www.wikidata.org/prop/>

select ?item  ?wditem ?itemLabel ?wid ?wditemlabel
where {
  ?item wdt:P17 wd:Q39.
  ?item wdt:P31 wd:Q5.
  optional {
    ?item wdt:P14 ?wid .
  }
  ?item rdfs:label ?itemLabel.   
  SERVICE <https://query.wikidata.org/sparql> {
    ?wditem xwdt:P27 xwd:Q258.
    ?wditem xwdt:P106 xwd:Q937857.
    ?wditem rdfs:label ?wditemlabel.
    filter(LANGMATCHES(LANG(?wditemlabel), "en")).
  }
  filter(contains(?wditemlabel, ?itemLabel))
}
group by ?item ?itemLabel ?wid ?wditem ?wditemlabel

The above works and matches items by their labels however:

1) I initially had filter(contains(?wditemlabel, ?itemLabel)) inside the SERVICE clause and it returned no results. But it seemed to work if I used a static string for one of the variables (e.g. filter(contains("test string", ?itemLabel))). Why would it work when comparing a variable and a string but not two variables?

2) I expected the query to work without the "group by" at the end. But it looks like without it, some sort of cross join/Cartesian product occurs and each item that is matched is repeated the total number of times (n * n). What part of the query is causing this?

Stanislav Kralin
  • 11,070
  • 4
  • 35
  • 58
Elfalem
  • 347
  • 3
  • 17

1 Answers1

1

Executing federated query, your local Blazegraph performs queries of this kind to Wikidata:

SELECT ?wditem ?wditemlabel
WHERE {
    ?wditem wdt:P27 wd:Q258.
    ?wditem wdt:P106 wd:Q937857.
    ?wditem rdfs:label ?wditemlabel.
    filter(LANGMATCHES(LANG(?wditemlabel), "en"))
    filter(contains(?wditemlabel, ?itemlabel))
}
VALUES () {
( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )  ( ) ( ) ( ) ( ) ( )
} # 100 values

As you can see, Blazegraph "forgets" to pass local bindings of ?itemLabel into VALUES — probably because ?itemLabel does not occur in remote triple patterns — but "thinks" that they were passed.

This bug causes both your problems:

  1. Try the above query on Wikidata (0 results)
  2. Try the above query on Wikidata without contains (82800 result instead of 828)

Workarounds

Force query execution order using hints:

select ?item ?wditem ?itemLabel ?wditemlabel
where {
  hint:Query hint:optimizer "None"
  SERVICE <https://query.wikidata.org/sparql> {
    ?wditem wdt:P27 wd:Q258.
    ?wditem wdt:P106 wd:Q937857.
    ?wditem rdfs:label ?wditemlabel.
    filter(lang(?wditemlabel)= "en").
  } 
  ?item wdt:P17 wd:Q39.
  ?item wdt:P31 wd:Q5.
  ?item rdfs:label ?itemLabel.
  filter(contains(?wditemlabel, ?itemLabel))
}

or

select ?item ?wditem ?itemLabel ?wditemlabel
where {
  ?item wdt:P17 wd:Q39.
  ?item wdt:P31 wd:Q5.
  ?item rdfs:label ?itemLabel.
  SERVICE <https://query.wikidata.org/sparql> {
    ?wditem wdt:P27 wd:Q258.
    ?wditem wdt:P106 wd:Q937857.
    ?wditem rdfs:label ?wditemlabel.
    filter(lang(?wditemlabel)= "en").
  }
  hint:Prior hint:runFirst true .
  filter(contains(?wditemlabel, ?itemLabel))
}

By the way, you could use DISTINCT instead of GROUP BY in your original query, or use additional local filtering, i. e. filter(lang(?itemLabel)='ast').

Comparison

In GraphDB, original query works well, but one should replace contains(?wditemlabel, ?itemLabel) with contains(str(?wditemlabel), str(?itemLabel)).

See also

Stanislav Kralin
  • 11,070
  • 4
  • 35
  • 58