How to know if a string is a proper name of person or a place name using DBpedia? - sparql

Question

I am using SPARQL query on DBpedia into a Prolog project and I have a doubt. I would know if a word is, most probably, a NAME OF A PERSON (something like: John, Mario) or a PLACE (like a city: Rome, London, New York).
I have implement the following two queries, the first gives me the number of persons having a specific name, and the second gives me the number of places having a specific name.
1) Query for a PERSON NAME:
select COUNT(?person) where {
?person a dbpedia-owl:Person .
{ ?person foaf:givenName "John"#en }
UNION
{ ?person foaf:surname "John"#en }
}
For the name John, I obtain the following output: callret-0: 7313, so I think that it has found 7313 instances for the proper name John. Is it right?
2) Query for a PLACE NAME:
select COUNT(?place) where {
?place a dbpedia-owl:Place .
{ ?x rdfs:label "John"#en }
}
The problem is that, as you can see in the previous “place” query, I have inserted John as parameter, which is not a place name but a proper name of persons, but I obtain the following strange result: callret-0: 81900104
The problem is that, in this way, if I compare the output of the previous two queries, it seems that John is a place and not a person name! This is not good for my scope; I have tried with other personal names and it always happens that the place query gives me a bigger output than the name query.
Why? What am I missing? Are there some errors in my queries? How can I solve it to have a correct result?

Answer 1

Actually, when I run the query you provided:
select COUNT(?place) where {
?place a dbpedia-owl:Place .
{ ?x rdfs:label "John"#en }
}
I get the result 93027312, not 81900104, but that does not really matter much. The strange results arise because ?x and ?place don't have to be bound to the same thing, so you are getting all the dbpedia-owl:Places and counting them, but the number of result rows is the number of dbpedia-owl:Place multiplied by the number of things with rdfs:label "John#en":
select COUNT(?place) where { ?place a dbpedia-owl:Place }
=> 646023
select COUNT(?x) where { ?x rdfs:label "John"#en }
=> 144
646023 × 144 = 93027312
If you actually ask for dbpedia-owl:Places that have the rdfs:label "John#en", you'll get no results:
select COUNT(?place) as ?numPlaces where {
?place a dbpedia-owl:Place ;
rdfs:label "John"#en .
}
SPARQL results
Also, you might consider using dbpprop:name instead of rdfs:label. Some results seem like they are more useful that way. For instance, let us find places called "Springfield". If we ask for places with that name we get no results:
select * where {
?place a dbpedia-owl:Place ;
rdfs:label "Springfield"#en .
}
SPARQL results
However, if we modify the query and use dbpprop:name, we get 17. Some of these are duplicates though, so you might have to do something else to remove duplicates. The point, though, is that dbpprop:name got some results, and rdfs:label didn't.
select * where {
?place a dbpedia-owl:Place ;
dbpprop:name "Springfield"#en .
}
SPARQL results
You can even use dbpprop:name in working with the names of persons, although it's not as useful, because the dbpprop:name value for most persons is their entire name. To find persons with the given name John using dbpprop:name requires a query like:
select * where {
?place a dbpedia-owl:Person ;
dbpprop:name ?name .
FILTER( STRSTARTS( str( ?name ), "John" ) )
}
(or you could use CONTAINS instead of STRSTARTS), but this becomes much more expensive, because it has to select all persons and their names, and then filter through that set. Being able to select persons based on a specific name (e.g., with foaf:givenName) is much more efficient.