chris mungall wrote:
I think both approaches are a little too XML-centric; fine for a few
use cases but in general the syntax obscures the declarative semantics
of the mapping which must be kept as perspicuous as possible. Why not
just use an RDF query language? This could be used to query over
generic DOM-as-RDF or relations-as-RDF mappings. This would seem to
leave open plenty of room for optimisation, scalability - and the data
sources could be queried directly (albeit slowly) rather than having to
refresh your data warehouse.
See also
http://scholar.google.com/url?sa=U&q=http://www.ics.forth.gr/isl/
publications/paperlink/Koffina.pdf
Chris - interesting link. As I mentioned earlier[1], I think that
describing mappings in a SW language is preferable, if for no other
reason than for uniformity. I agree that if you are going to perform the
transformation from semantic-less data in RDF (syntactic data types
only, with, say, column names from tabular data) to semantically tagged
data (the same data but with OWL types), you might as well do it in a
RDF query language (preferably SPARQL). We were about to do just that
with a SeRQL construct statement but found that we could avoid
rebuilding the entire RDF graph of the data by simply declaring
'data-level' properties to be subproperties of OWL types in RDFS
(allowing the auto-inferencing of the Sesame RDFS repository to take
care of rdfs:type). We were subsequently able to issue data integration
queries in our own OWL terms [see attached image]. This seems to work
well for "equivalence" mappings. Is there a best practice for this?
I think that there are some features of translation to RDF that should
be carefully noted. In particular, I think that it is important to
distinguish between translations that are direct mappings, where terms
are simply translated to RDF syntax ala the YeastHub article[2] and
those that include semantic mappings, such as the RDFS subproperty
"trick" mentioned above. I agree with Peter Mork where he pointed out in
[3] that the road to semantic annotation is best achieved in two
explicit transformations: syntactic and semantic.
-scott
[1]http://lists.w3.org/Archives/Public/public-semweb-lifesci/2006Feb/0171.html
[2]http://bioinformatics.oxfordjournals.org/cgi/reprint/21/suppl_1/i85
[3]http://lists.w3.org/Archives/Public/public-semweb-lifesci/2006Feb/0172.html