W3C home > Mailing lists > Public > www-rdf-interest@w3.org > January 2003

Re: Efficient context mechanisms

From: Richard H. McCullough <rhm@cdepot.net>
Date: Tue, 14 Jan 2003 11:37:54 -0800
Message-ID: <001001c2bc04$86859e30$bf7ba8c0@rhm8200>
To: "Dave Reynolds" <der@hplb.hpl.hp.com>, "Bob MacGregor" <macgregor@ISI.EDU>
Cc: <www-rdf-interest@w3.org>
FYI: 
in MKR/KE

1. time context
    at time=t1,view=c1 { statement1 }
    at time=t2,view=c1 { statement2 }
is one view context with two events (assuming statement1,statement2 are action statements).

2. multiple contexts for one statement
    at view=[c1, c2, c3] { statement1 }
is a legal proposition.
Multiple contexts are not currently implemented in KE, because I didn't anticipate the need for it.  
However, a few minutes coding will remedy that situation.
============ 
Dick McCullough 
knowledge := man do identify od existent done
knowledge haspart proposition list

  ----- Original Message ----- 
  From: Bob MacGregor 
  To: Dave Reynolds 
  Cc: www-rdf-interest@w3.org ; tar@ISI.EDU ; hans@ISI.EDU 
  Sent: Tuesday, January 14, 2003 9:45 AM
  Subject: Re: Efficient context mechanisms

  Dave,
  > > [Discussion on contexts]
  >
  > > If you ever decide to get serious about contexts, it would probably be
  > > worth your
  > > while to assign someone the task of deciphering our context algorithm (or
  > > Austin's).
  > > It may be overkill for RDF, but some of the lessons learned probably 
  > apply to
  > > simpler context systems.
  >
  >I agree. A serious context mechanism is not on our short term priority 
  >list but
  >it's an interesting area to think about in the longer term. The open world
  >assumption behind the semantic web makes it likely that some applications will
  >require non-monotonic reasoning, some forms of which would benefit from 
  >such an
  >efficient context mechanism.

  We have an application that is tracking objects across time.  We seem
  to need a context per time frame (or a model per time frame).  So contexts
  can arise for very simple datasets.

  >What is the space complexity of the algorithm like?

  The contexts are organized into a hierarchy, and numbered so
  that a parent's number is always smaller than the numbers of its
  descendants.  Some sort of data structure is needed that can quickly
  determine, given two contexts C1 and C2, whether C1 inherits C2.
  One could imagine that the space requirements for this would be
  superlinear in the worst case, but they should be linear (in the number
  of contexts) for the average case.

  Each statement/triple points to a list of contexts where it has
  appeared within an assertion. If a triple [S P O] is asserted to be true in
  contexts C1, C2, and C3, then there is a triple object that points
  to a list containing C1, C2, and C3 (ordered according to the
  numbers they are assigned, largest number first).

  If we were using a quad representation, then memory is linear in
  the number of quads.  However, wrt quads, I'm not sure our notion is
  in sych with the statings crowd.  Suppose the quad form is

  [C1 S P O]
  [C2 S P O]
  [C3 S P O]

  While we conceptualize this as one triple [S P O], asserted
  in three different contexts, the statings crowd seems two think of
  these quads as three different statings that are only loosely
  related.

  What the semantics of the above is can't really be resolved unless we
  decide on what a quad is (for example, must one of its arguments always
  be a context?) and what a context is (e.g., is a context a set of
  assertions or a set of statings?).

  Cheers, Bob
Received on Tuesday, 14 January 2003 14:38:46 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:51:57 GMT