Re: Multidimensional vs. One-dimensional info WAS Re: [xtm-wg] A challenge on "the graph"

["Thomas B. Passin" <tpassin@home.com>]

Sam Hunting asked -"

> [tom p]
> > Whatever do you mean? They become values of nodes, possibly in their
> > own nodes that are linked to parent nodes.  The real issue in going
> > from brackets to the in-memory (or abstract model) is that the pointy
> > stuff is serialized and the full model is interconnected and not easy
> > to serialize.
>
> Tom--
>
> Maybe you could expand a little on why the full model has to be
> interconnected as it is? This idea seems so basic to me that I'm having
> a hard time backing off it enough to put it into words. Many thanks.
>
Yes.  First, consider a (non-topic map) xml document that has no IDs and no
references to IDs.  Each child element is "contained" by its parent.  A
nice, simple, clean, serial structure.  Now it's easy to model this document
as a series of objects (not even "nodes, really"), and each object has a
list of children, each of which in turn can have its own list of children.
Attributes can turn into another list or hash.

Such a memory structure does not have to be represented or thought of as a
graph, and is easy to work with.  It's the objects have properties which may
have properties model.  It's duck soup in python, for example (I mean, easy
to program).

But a Topic Map is filled with items - topics, association roles, and so
on - that refer to other items.  Topics are instances of other topics, and
so on.  These are the "interconnections" that I mentioned.  The final
structure, however it may be implemented in the computer's memory, is a
densely interconnected (I don't mean "dense" in a mathematical sense,
there's probably a technical meaning for the term) arrangement of the
"items".  It can't be represented as a tree because there are multiple
routes between many of the items.  If it's going to be a graph at all, it's
probably multi-rooted, and possibly cyclic as well (I haven't thought about
this angle much yet)

An XML document is a serialized version of a tree - a single-rooted, acyclic
graph.  To serialize the non-tree, that is, to work around the limitation of
non-cyclic, single-rootedness, IDs and IDrefs are used to augment the
"containment" of child elements within children.  Once it's in memory, you
could have all these items linked up without any explicit ID at all, they're
mainly needed for serializing.  If you doubt this, just think of a linked
list - you can work with each member of the list with no explicit notion of
its memory address or object id.  The same would be true of an in-memory
representation of a topic map.  The computer would, of course,  use
references that it assigned, but there would be no need for the ID
attributes per se.

In a complicated topic map, the interconnections between items will be more
numerous than the direct containment.  For example, an element may be
referred to as a parent type 1000 times, while its directly contained
children - its names and occurrances - will be few by comparison, perhaps as
few as one or two.

In fact, it's these links that will bring the primary value of topic maps.
This suggests that the emphasis in the models should be on the links.  This
in turn suggests the use of graphs.  What other approach is there that is
particularly strong in its approach to links?

I see I got a bit long-winded here.  Sam, did this satisfy your question?

Regards,

Tom P




------------------------ Yahoo! Groups Sponsor ---------------------~-~>
Find software faster. Search more than 20,000
software solutions on KnowledgeStorm. Register
now and get started.
http://us.click.yahoo.com/556S5A/RNSCAA/2h4EAA/2n6YlB/TM
---------------------------------------------------------------------_->

To Post a message, send it to:   xtm-wg@eGroups.com

To Unsubscribe, send a blank message to: xtm-wg-unsubscribe@eGroups.com 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/