- From: Jeremy Carroll <jjc@hplb.hpl.hp.com>
- Date: Sun, 29 Jun 2003 21:39:39 +0100
- CC: Charles McCathieNevile <charles@w3.org>, rdf-i <www-rdf-interest@w3.org>
(Although I've not read the whole thread) Yes - the ordering of the subjects depends on any bnode IDs and the bnode IDs are arbitrary. The most obvious way to assign them is in an increasing numeric sequence, but then which comes first - the ordering of the triples (which depends on the bnode IDs) or the bnodeIDs which depend on the ordering of the triples. All in all, any solution that allows RDF canonicalization in general, must permit the comparison between two arbitrary rdf graphs. Since arbitrary unlabelled undirected graphs can be embedded in RDF, two such graphs can be comapred. This is known as the graph isomorphism GI problem, so any complete solution for RDF canonicalization solves GI for which no polynomial time solutions are known. I have explored this issue in detail, and believe, and argue in a paper that I will present in ISWC2003 (I will try and get a pdf available soon), that despite this a practical solution is achievable. More tomorrow. Jeremy
Received on Sunday, 29 June 2003 16:40:59 UTC