W3C home > Mailing lists > Public > uri@w3.org > April 2002

RE: silly question about rdf:about

From: Graham Klyne <GK@ninebynine.org>
Date: Fri, 12 Apr 2002 11:08:01 +0100
Message-Id: <5.1.0.14.2.20020412105908.03b24ec0@joy.songbird.com>
To: <LMM@acm.org>, "Joshua Allen" <joshuaa@microsoft.com>
Cc: <uri@w3.org>
Hi,

I glanced through this and an uncomfortable thought occurred to me.  I 
think it's fairly clear that given two duri's that differ only in their 
date element, they may or may not denote the same thing depending upon 
other factors.  Thus, given Joshua's employment record, one might be able 
to infer that
   urn:tdb:2001:http://www.netcrucible.com/people/joshuaa
   urn:tdb:2002:http://www.netcrucible.com/people/joshuaa
do or do not denote the same person.

But, and this is where I get uncomfortable, to perform such inference one 
needs to look _inside_ the URI syntax, which seems to fly in the face of 
universality/uniformity of URIs used in this way.  In RDF/N3-speak, 
something more like this would seem to be more appropriate:

   [a :Person] :describedBy [a :Description ;
                             :dateAvailable "2001" ;
                             :homePage 
<http://www.netcrucible.com/people/joshuaa> ] .

Now, given an employment record, it seems much more comfortable to do the 
inference to determine if that above describes the same person as this:

   [a :Person] :describedBy [a :Description ;
                             :dateAvailable "2002" ;
                             :homePage 
<http://www.netcrucible.com/people/joshuaa> ] .

#g
--

At 09:24 PM 4/11/02 -0700, Joshua Allen wrote:
> > If you leave netcrucible, someone else named Joshua Aardvaark joins
>and
> > puts up http://www.netcrucible.com/people/joshuaa, it's a different
> > person.
>
>Right, that is what my UAW example was showing.  I'm simply pointing out
>that a date is not terribly useful for disambiguation (well, actually I
>don't know exactly what the date is useful for).
>
> > Maybe it's reasonable -- and maybe not -- to treat two tdbs which
>differ
> > by a small date range to be 'the same'.
>
>I can accept this.
>
> > Can I post your comments & my reply on the 'uri@w3.org' list?
>
>Done :-)
>
>Thanks,
>Joshua
>
> > -----Original Message-----
> > From: Larry Masinter [mailto:LMM@acm.org]
> > Sent: Thursday, April 11, 2002 5:19 PM
> > To: Joshua Allen
> >
> > I think that the problem of persistence of identity is really
>difficult --
> > it's one of those problems like "is the morning star the same as the
> > evening star", because, after all, they appear at different times. And
> > with 'tdb', there are two kinds of ways in which meaning can vary over
> > time: (a) the web resource itself could change (the corresponding
>urn:duri
> > might yield a different document), and (b) the understanding of what's
> > being described might change.
> >
> > If you leave netcrucible, someone else named Joshua Aardvaark joins
>and
> > puts up  http://www.netcrucible.com/people/joshuaa, it's a different
> > person.
> >
> > I don't think you can assume there are unique identifiers for people
>(or
> > anything else), unless you choose a narrower domain of identification
>that
> > tries to assure that each 'thing' only gets one identity, e.g.,
>identify
> > you by your SSN or some other country's national identity number.
> >
> > In the end, a meta-indexer has to use heuristics to decide whether two
> > identified resources are the 'same' for the purpose of the indexer's
>index.
> > Maybe it's reasonable -- and maybe not -- to treat two tdbs which
>differ
> > by a small date range to be 'the same'.
> >
> > I think this is completely orthogonal to the issue of media
>representation,
> > which includes a wide range of variability, not just MIME type.
> >
> > I don't think 'tdb' or 'duri' are useful in situations where there is
>a
> > 'client' and 'servers' and you're trying to do content negotiation. It
> > doesn't really help with that.
> >
> > Can I post your comments & my reply on the 'uri@w3.org' list?
> >
> >
> >
> >       -----Original Message-----
> >       From: Joshua Allen [mailto:joshuaa@microsoft.com]
> >       Sent: Monday, April 08, 2002 11:49 PM
> >       To: LMM@acm.org
> >       Subject: RE: silly question about rdf:about
> >
> >
> >       Yeah, I read this last night when Aaron posted about it on
>#rdfig.
> > I like the idea of a tdb URI scheme, but I am uneasy about putting a
>date
> > into a tdb.  The section 5.6 comment applies more to duri than tdb;
>that
> > is, one would expect that the thing being described would still be the
> > same thing after a period of time.  That is, the two URIs:
> >
> >       urn:tdb:2001:http://www.netcrucible.com/people/joshuaa
> >       urn:tdb:2002:http://www.netcrucible.com/people/joshuaa
> >
> >       are different URIs, but most likely identify the same "thing".
> > The fact that the URIs are different would complicate (but not
>prevent)
> > attempts to index assertions about a particular "thing".
> >
> >       An example of where the date would be useful would be something
>like:
> >
> >       urn:tdb:1996:http://www.uaw.org/officers/president
> >       urn:tdb:2002:http://www.uaw.org/officers/president
> >
> >       Which do describe different things.  But even then, a metadata
> > indexer is left with two choices -- you either index all data for each
> > distinct date as if they describe different "things", or index them as
>if
> > they all describe the same thing.  Either option is unlikely to give
> > accurate metadata aggregation.  In other words, the date is relatively
> > useless to disambiguate the tdb, so other techniques would have to be
>used.
> >
> >       So, I believe that the tdb scheme would be useful to divert
>people
> > who insist on saying that an HTTP URI represents more than just a
>document.
> > For that alone, it would be nice.  There would of course still be the
>"One
> > HTTP URL can represent many different resources by use of a mime-type"
> > arguments, so those would need to be put to rest somehow.  And beyond
>that,
> > I am still trying to think of places where I would want to use the
>duri
> > (non-tdb).  At first glance, it seems appealing; a sort of
>metadata-safe
> > way to do HTTP etags.  But etags are most useful when maintained by
>the
> > server, and not by the client -- I think the same would be true of the
> > datepart in tdb and duri -- but in the case of tdb and duri, if a
>server
> > is going to go through the trouble of updating the datepart of the URN
> > every time that a resource change (or change of president at the UAW)
> > requires, then the server might as well just assign a new URN.  So I
>am
> > still trying to get the use cases.
> >
> >
> >
> >               -----Original Message-----
> >               From: Larry Masinter [mailto:LMM@acm.org]
> >               Sent: Mon 4/8/2002 9:01 PM
> >               To: Joshua Allen
> >               Cc:
> >               Subject: RE: silly question about rdf:about
> >
> >
> >
> >               > The most important thing to me, by far, is just to
>have a
> > decision and
> >               > get rid of the uncertainty.  I wish someone would just
> > declare that
> >               > "this is how URI's for identity work, and all other
>ways are
> >               > discouraged."
> >
> >               How about "tdb" in
> > http://larry.masinter.net/duri.[html,txt,xml] ?
> >
> >

-------------------
Graham Klyne
<GK@NineByNine.org>
Received on Friday, 12 April 2002 06:30:07 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Thursday, 13 January 2011 12:15:30 GMT