- From: Enrico Franconi <franconi@inf.unibz.it>
- Date: Wed, 12 Dec 2012 21:35:16 +0100
- To: David Booth <david@dbooth.org>
- Cc: Pat Hayes <phayes@ihmc.us>, semantic-web <semantic-web@w3.org>
On 12 Dec 2012, at 20:09, David Booth <david@dbooth.org> wrote: > But according to > http://web.ing.puc.cl/~marenas/publications/iswc11.pdf > the vast majority of RDF documents (over 98% of their samples) use blank > nodes in non-problematic ways. (I.e., they contain no blank node > cycles, and thus do not cause the graph isomorphism complexity problem.) > At present the many applications that process RDF have to pay for the > sins of those (very) few RDF graphs that use blank nodes in problematic > ways. > > Actually, it would be interesting to examine whether those <2% of graphs > that did have blank node cycles really needed them. My suspicion is > that the authors could have simply minted a few URIs to break those > blank node cycles and turn them into non-problematic blank node trees. > In the nearly 4 million RDF documents Mallea, Arenas, Hogan, and > Polleres examined, the maximum blank node treewidth they found was 7, > which I think (though a graph theory expert would have to confirm) that > only 6 URIs would have to have been minted to turn it into a tree. My suspicion (also compatible with the experimental data) is that people are using RDF to naïvely implement an "object model". I believe that RDF is not used in its full power simply because people are not used to model directly with a relation-based language, where bnode-cycles would be more natural. Is this good? Probably not, since if we actually need an object model then we could go back to Corba :-) [just kidding]. Technically speaking, the idea to simplify RDF by disallowing bnode-cycles is appealing. cheers --e.
Received on Wednesday, 12 December 2012 20:35:51 UTC