- From: Adrian Walker <adriandwalker@gmail.com>
- Date: Tue, 31 Jul 2007 16:25:10 -0400
- To: "Chris Bizer" <chris@bizer.de>
- Cc: "Hugh Glaser" <hg@ecs.soton.ac.uk>, "Pat Hayes" <phayes@ihmc.us>, "Tim Berners-Lee" <timbl@w3.org>, semantic-web@w3.org
- Message-ID: <1e89d6a40707311325t65f5950bm90e225eebd038801@mail.gmail.com>
Hi Chris -
You wrote...
I think it is a bit naive to throw lots of RDF data from the Web
straight into a single RDF model and then wonder that reasoning over this
data leads to unintended consequences.
...
In the light of the current Semantic Web layer cake discussion, I have been
wondering for years why the trust layer is up that far in the layer cake. It
is obvious that you will only get junk if you try to reason over data from
the web before applying some heuristics to determine trustworthiness and
filter out low quality information.
Absolutely! But would someone bet their business, or the outcome of a
military operation, on heuristics? Consider that the heuristics would be
relatively static, while the RDF data from the web could be changing fast.
There's a complementary way of building Trust, as follows. Make sure that
the application logic is linked computationally to an English description of
what the application author(s) intended. On demand, extract a proof tree
for for a derived answer, and map it computationally to an English
explanation that non-techies can read. (It can also explain what the
heuristics did.)
Here's a simple example of how this kind of approach can work:
www.reengineeringllc.com/demo_agents/RDFQueryLangComparison1.agent
Also, here's an example to show that the same approach can work over SQL,
and therefore over SPARQL too:
www.reengineeringllc.com/demo_agents/Oil-IndustrySupplyChain1MySql1.agent
You can run the examples, and see the explanations of the results, by
logging into the system at the same site.
BTW, the discussion so far focuses on wrong answers. You will also Trust a
system more if you believe that it does not miss answers. Abductive style
explanations (as in the online system below) can help with that aspect of
Trust too.
Perhaps the Trust layer in the Cake should be labelled "Explanation and
Trust" ? Or "Proof" should be relabelled "Explanation" ?
Cheers, -- Adrian
Internet Business Logic
A Wiki for Executable Open Vocabulary English
Online at www.reengineeringllc.com Shared use is free
Adrian Walker
Reengineering
Received on Tuesday, 31 July 2007 20:25:14 UTC