- From: Chimezie Ogbuji <chimezie@gmail.com>
- Date: Fri, 23 Nov 2007 17:20:51 -0500
- To: sean@miscoranda.com, semantic-web@w3.org
> One of the biggest Semantic Web questions people are asking right now is: when a Semantic Web User Agent gets a document, how many normative ways of getting triples from it are there? Yes, a question I've been asking repeatedly over the past year or so with very little useful response or any indication that there is any coherent consensus. This does not bode well if we wish the Semantic Web to indeed be a framework for autonomous agents. GRDDL, RDFa, etc.. get us much of the way there, but a more comprehensive "agent contract" is needed. > Or, from the other direction: how many triples is the author asserting in some document? This is a bit divergent since it relies on many things including (but not limited to): a linkage contract (which RDF predicates did the author use to indicate that the linked graphs are normatively included), an entailment regime (we still don't have a solid notion for this), .. I'm sure I'm missing other factors .. > I'm proposing some kind of work on conformance levels for Semantic Web User Agents, such that when someone says "how many triples are in $uri", we can answer confidently "a Class 2 Semantic Web User Agent will return 53 triples"; or perhaps not *that* abstract, but along those lines. Given the scattered nature of this space, I think this would be very prudent. > The aim is for document producers to know how many UAs out there support the format that they're using, and to give some kind of regularity to what is currently a bundle of ad hoc solutions. Absolutely.. > As a bit of context, here are some people who've been thinking about this specifically from an engineering point of view: By the way, I've since dusted off a set of diagrams i drew earlier this year while trying to tackle this very problem. I started off writing a generic semantic web agent "module" which plugs into a 'vanilla' RDF processing / inference library thinking it should be a no brainer and quickly found out how significantly tangled the landscape is in this regard. I've put together a wiki with some of my thoughts (which also pull from earlier [http://redfoot.net Redfoot] ideas): http://code.google.com/p/python-dlp/wiki/Agentem > There's also been a lot of discussion about Xiaoshu Wang's paper on kinda the same issues, but I think that's a distraction so I won't bother to link to it. I think those discussions (though related) are more about web architecture "semantics" than a standard framework for semantic web agents. I consider the two separate and conversation that suggests that the overlap between the two is significant is part of the problem here. [snip] > What a huge tax on Semantic Web User Agents that also have to be conforming GRDDL agents and conforming RDF/XML agents and so on! Agreed. A discrete set of SWUA capabilities would go along way in doing some damage control on the tax which is mostly because compliance (for each framework) is defined in terms of an isolated, all-or-nothing set of requirements. > GRDDL in particular is a tricky case because the conformance is left open but it recommends that you implement *all* of the underlying processing mechanisms and then XSLT 1.0 as the main transformation language; but this particular conformance class doesn't have a name, it's just represented in the GRDDL Test Cases REC. Well, let's start the conversation you are suggesting by giving this class of GRDDL Aware Agents a name :) - In the FSM diagrams on the Agentem wiki above, I tried to indicate where such a set of conformance classes might fit in the larger picture. [[[ If you really want to take this to its logical conclusion, it would be nice to have a vocabulary for describing the capabilities of Semantic Web user agents to consume various documents, a writeup of the heuristics that they ought to use, and a kind of extra layer of conformance levels for Semantic Web user agent authors to try to meet. "Don't wanna support all of GRDDL? Here are a few common subsets that are well deployed." This should be based on some level of description, looking to see what kinds of documents people are actually using, and prescription, what kinds would be good to produce especially in future when things like RDFa go to rec. ]]] - Best Practices Issue: RDF Format Discovery http://lists.w3.org/Archives/Public/public-swd-wg/2007Nov/0056 Sounds like a (much needed) first step in the right direction. I think it would be a shame if none of what you suggest is considered simply as a result of it being out of scope for the forums you pinged. The vision of a web of data where agents can behave efficiently on our behalf will not happen without a framework which contributes some minimal amount of coherence and semi-determinism. My $0.02 -- Chimezie
Received on Friday, 23 November 2007 22:27:56 UTC