W3C home > Mailing lists > Public > www-rdf-interest@w3.org > February 2001

Re: About using distances or probabilities

From: Seth Russell <seth@robustai.net>
Date: Fri, 09 Feb 2001 13:38:24 -0800
Message-ID: <3A846350.76E6E8A3@robustai.net>
To: David Allsopp <dallsopp@signal.dera.gov.uk>
CC: www-rdf-interest@w3.org
David Allsopp wrote:

> Seth Russell wrote:
> >
> > You could hang the probabilites off the property nodes:
> >
> > :Computer :pal20/30  :Printer.
> > :pal20/30  :probability "20"; :inverseProbability "30"; a :Property.
>
> Presumably this requires a unique Property instance for each different
> link between nodes, for example if I have
>
> John--believes-->SkyIsBlue
> John--believes-->MagicPixies
>
> and want to assign 99% confidence to the first one and 1% to the second,
> I'd need a 'believes99' property and a 'believes01' property

Yes.  But I think we need to distinguish between what happens inside a
implementation and what that application communicates to the external world in an
RDF expression.   If I were to design an internal implementation that had fuzzy
probabilities on arcs, I would put the probabilities inside of the arcs and store
them in some kind of ntuples.  But if that data needed  to be communicated to the
external world in RDF, then hanging the probabilities off the property nodes is
one option ...

> (or reify both statements and attach the confidence level to those) ?

and I suppose that would be another.

Seth
Received on Friday, 9 February 2001 16:29:48 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:51:47 GMT