Re: ISSUE-9 Another question about Generate Blank Nodes

* Souripriya Das <SOURIPRIYA.DAS@oracle.com> [2011-02-01 18:55-0800]
> Let me use the same test to illustrate a problem with use of bNodes vs. IRIs:
> http://www.w3.org/2001/sw/rdb2rdf/wiki/R2RML_Test_Cases_v1#2duplicates0nulls
> 
> we have
> ┌┤IOUs├─┬───────┬────────┐
> │ fname │ lname │ amount │
> │ Bob   │ Smith │     30 │
> │ Sue   │ Jones │     20 │
> │ Bob   │ Smith │     30 │
> └───────┴───────┴────────┘
> Here are the generated triples, without showing the actual destination graph:
> 
> Using IRIs:
> <IOUs/fname=Bob,lname=Smith,amount=30:1#_> <IOUs#fname> "Bob" ;
>                                            <IOUs#lname> "Smith" ;
>                                            <IOUs#amount> 30.0 .
> <IOUs/fname=Sue,lname=Jones,amount=20:1#_> <IOUs#fname> "Sue" ;
>                                            <IOUs#lname> "Jones" ;
>                                            <IOUs#amount> 20.0 .
> <IOUs/fname=Bob,lname=Smith,amount=30:2#_> <IOUs#fname> "Bob" ;
>                                            <IOUs#lname> "Smith" ;
>                                            <IOUs#amount> 30.0 .
> Using bNodes:
> _:b1 rdf:type ex:IOUs;
>          table:fname "Bob";
>          table:lname "Smith"; 
>          table:amount "30"^^xsd:integer.
> 
> _:b2 rdf:type ex:IOUs;
>          table:fname "Sue";
>          table:lname "Jones"; 
>          table:amount "20"^^xsd:integer.
> 
> _:b3 rdf:type ex:IOUs;
>          table:fname "Bob";
>          table:lname "Smith"; 
>          table:amount "30"^^xsd:integer.
> 
> Now suppose that the triples generated from the table go to more than one named graphs:
> - Graph <G1> gets the triples generated from the columns "fname" and "lname"
> - Graph <G2> gets the triples generated from the column "amount"
> - (this is not important here, but) assume that both graphs get the rdf:type triples
> 
> The following set of triples clearly represents a different RDF data than the sets shown above due to graph-local scope of bNodes:
> 
> <G1> 
> {
> _:b1 rdf:type ex:IOUs;
>          table:fname "Bob";
>          table:lname "Smith" . 
> _:b2 rdf:type ex:IOUs;
>          table:fname "Sue";
>          table:lname "Jones" .
> _:b3 rdf:type ex:IOUs;
>          table:fname "Bob";
>          table:lname "Smith" .
> }
> 
> <G2> 
> {
> _:b1 rdf:type ex:IOUs;
>          table:amount "30"^^xsd:integer.
> _:b2 rdf:type ex:IOUs;
>          table:amount "20"^^xsd:integer.
> _:b3 rdf:type ex:IOUs;
>          table:amount "30"^^xsd:integer.
> }
> 
> It is possible to introduce IRIs as go-between and also use owl:sameAs to get back to the original semantics of the relational data, but it gets complicated.

I would expect that you'd need to use domain information to create
some persistent identifiers anyways in order to be complete in the
face of queries explicitly mentioning the subjects of any of those
triples. (Queries which don't include any of those nodes are
unaffected by whether they are IRIs or blank nodes.)

It's an interesting problem; convenience vs. completeness.


> Thanks,
> - Souri.
> 
> ----- Original Message -----
> From: eric@w3.org
> To: ashok.malhotra@oracle.com
> Cc: auer@informatik.uni-leipzig.de, juanfederico@gmail.com, public-rdb2rdf-wg@w3.org
> Sent: Tuesday, February 1, 2011 8:18:44 PM GMT -05:00 US/Canada Eastern
> Subject: Re: ISSUE-9 Another question about Generate Blank Nodes
> 
> * ashok malhotra <ashok.malhotra@oracle.com> [2011-02-01 15:33-0800]
> > Let me see if I understand the problem.
> > 
> > Suppose we have a table with no primary key and many columns.
> > A triple would be generated for each column in the table and all the triples
> > for a row would be anchored by a blank node.  Is this correct?
> 
> Yes, for
> http://www.w3.org/2001/sw/rdb2rdf/wiki/R2RML_Test_Cases_v1#2duplicates0nulls
> we have
> ┌┤IOUs├─┬───────┬────────┐
> │ fname │ lname │ amount │
> │ Bob   │ Smith │     30 │
> │ Sue   │ Jones │     20 │
> │ Bob   │ Smith │     30 │
> └───────┴───────┴────────┘
> The direct graph Sören proposes would look like:
> 
> <IOUs/fname=Bob,lname=Smith,amount=30:1#_> <IOUs#fname> "Bob" ;
>                                            <IOUs#fname> "Smith" ;
>                                            <IOUs#fname> 30.0 .
> <IOUs/fname=Sue,lname=Jones,amount=20:1#_> <IOUs#fname> "Sue" ;
>                                            <IOUs#fname> "Jones" ;
>                                            <IOUs#fname> 20.0 .
> <IOUs/fname=Bob,lname=Smith,amount=30:2#_> <IOUs#fname> "Bob" ;
>                                            <IOUs#fname> "Smith" ;
>                                            <IOUs#fname> 30.0 .
> 
> 
> > So, the problem is that we have a blank node anchoring the triples for each
> > row but, really, the blank nodes for each row represent different entities.  Is this correct?
> 
> Yes, with the caveat that they represent different rows in the
> database. The extent to which these rows represent different entities
> is a matter of database modeling, to be kept in mind when formulating
> queries.
> 
> 
> > If so, then I'm with Soeren.  We can improve the details of his solution but his
> > direction seems right.
> 
> I believe a motivation is to be linked-data-friendly; that is to
> identify things with URLs and to serve them when asked, as in
>   GET <IOUs/fname=Bob,lname=Smith,amount=30:2#_> HTTP/1.0
> 
> The thing we specifically don't want to do is to give the world and
> identifier which we have no way of resolving, either in response to
> a GET, or in response to a SPARQL query e.g.:
>   ASK { <IOUs/fname=Bob,lname=Smith,amount=30:2#_> <IOUs#fname> 30.0 }
> 
> If I use an Oracle rownum to tweak one of a set of identical rows, or
> if the rows are not identical and I use that to tweak the identifier,
> I can no longer honor queries about this row, even though the row
> still exists. I believe it is easier to honor the LD requirements if
> we tie the RDF identifiers directly to the SQL identifiers, and don't
> advertise identifiers otherwise.
> 
> 
> > All the best, Ashok
> > 
> > On 2/1/2011 2:11 PM, Sören Auer wrote:
> > >Hi all,
> > >
> > >In todays telco several people (including Souri and me) supported the idea to abandon the use of blank notes. Is there any fundamental reason (beside philosopical views) to use blank nodes?
> > >If not I suggest we just generate IRIs for all resources. Of course this does not yet solve the problem of how they should be created, but we could follow the following strategy:
> > >
> > >* if there is a candidate key use the candidate key,
> > >* if there is no candidate key, but an internal row identifier (e.g. Virtuoso has such one always) use this row identifier,
> > >* if nether one exists, generate an identifier using a hash function over all values of the row + an incremented counter in case duplicate rows exist
> > >
> > >Wouldn't this be a simple and effective solution to the problem?
> > >
> > >Best,
> > >
> > >Sören
> > >
> 
> -- 
> -ericP
> 
> 

-- 
-ericP

Received on Wednesday, 2 February 2011 06:15:01 UTC