I'm a newcomer here, so it's possible that this has been discussed to
death and I simply haven't found the relevant topics (although I
_have_ looked).  Bearing that in mind:

I ran across the following statement on the SMW Help pages (in
particular, < http://semantic-mediawiki.org/wiki/Help:RDF_export#Categories
>):

"There are many usages of MediaWiki categories that conflict with
these semantics. For example, the article Urban decay might be in
category Cities, but it is not a city. And Category:City museums might
be in category Cities, but city museums are not cities."

This strikes me as being a potentially serious problem, especially
when it comes to using an external ontology manager to do further
analysis on the appropriate SMW articles.  Conversely, it was
indicated elsewhere that these conflicting usages are actually quite
desirable as far as wikis are concerned, since categories are supposed
to group together articles of common interest.

>From the above, it seems to me that Categories are _not_ the
appropriate equivalent to OWL/RDF classes.  That is, the presence of a
'[[Category:xxx]]' annotation in an article should not automatically
map to an rdf:type element; nor should its presence in a Category page
automatically map to an rdfs:subClassOf element.  Bear with me,
please.

OTOH, I do agree that whatever annotation does map to rdf:type and
rdfs:subClassOf should be something that causes the article or
category in question to be added to a category.

Here's what I'm proposing:

1. Add a new Property Type: namely, Type:Category.  Like Type:Page,
Properties of Type:Category define a relation between two pages.
Unlike Type:Page, the page being pointed to is always a Category; and
including this Property on a page implicitly categorizes it.

By annotating pages with properties of this type, it becomes possible
to distinguish between different reasons for including the page in the
category.

2. Add a new special property: 'Class'.  Property:Class has type
Category; additionally, only categorization by means of Property:Class
would map to rdf:type and rdfs:subClassOf; all other means of
categorization (including the traditional '[[Category:foo]]'
annotation) would map to owl:ObjectProperty.

I'd also add a special property 'Category'.  Categorizing through it
would be identical to categorizing by the traditional mechanism: that
is, '[[Category::foo]]' would be completely interchangeable with
'[[Category:foo]]'.  But this isn't strictly necessary; it's merely
one way that you might ensure that you have a way of mapping
'[[Category:foo]]' to an owl:ObjectProperty element.

--

The main downside that I see to this proposal is that semantic wikis
that adopt it would have to be thoroughly reviewed in terms of their
category annotations.  OTOH, the only alternatives that I see would be
for a semantic wiki to insist that Categories only be used to
represent 'is-a' relationships, or for the semantic wiki to ignore the
issue.  (Technically, you could reverse my proposal by providing a
special Property that categorizes the page without establishing a
class/instance-or-subclass relationship when exported; but this is
essentially the same solution, changing only the "burden of proof", as
it were.)  It's possible for either of these to be viable solutions: a
wiki that's built entirely around a given vocabulary can get away with
mandating that all articles in a category must be instances of its
class, and an open-ended wiki such as Wikipedia may not encounter this
issue often enough to warrant the amount of work that would be needed
to fix the problem.

With this in mind, it's entirely possible that I'm proposing an SMW
Extension, rather than an upgrade.  That way, those wikis that want to
implement this could, while those that don't want to bother with it
don't have to.

Thoughts?

-- 
Jonathan "Dataweaver" Lang

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Semediawiki-devel mailing list
Semediawiki-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/semediawiki-devel

Reply via email to