Re: Spatial datatypes and RDF/OWL

Hi John, Steve, all:

Our group at BBN has been doing some work in this space for a while. 

We have a prototype for combined RDF and spatial storage, a paper on the 
approach is available at [1].

Our preference is to have the knowledgebase recognize spatial predicate 
triples and process them appropriately, rather than build functions into 
SPARQL's filter clauses.  This way, SPARQL parsers /engines need not be 
extended, all of it can be done at a lower level.

Your query would be written like:

SELECT ?a ?b

WHERE{
    ?a hasGeometry ?g1 .
    ?b hasGeometry ?g2 .
    ?g1 touches ?g2 .
}

So it accomplishes the same thing.  The same works for data insertion; 
things that look like points or polygons are detected and indexed.
Another nice thing about this approach is that these relationships can 
be asserted in an RDF document, for instance if they were being 
transferred to a KB that couldn't do spatial processing. 

Right now, the knowledgebase looks for these predicates explicitly; 
something like triggers might be useful for generalizing this.

One change that would improve our current prototype would be to replace 
the memory-based spatial index with a persistent one.  With the 
in-memory spatial index, there is a clear cap as to the number of 
spatial entites it can handle.  I actually chatted with Cathy Dolbear 
and Glen Hart a bit about this before we'd done any of the work. 

Thanks,
--Dave--

[1] http://iswc2007.semanticweb.org/papers/785.pdf

Received on Monday, 14 April 2008 18:01:36 UTC