Hello again.
I just had a descussion with JeLuF on #mediawiki about the concept of
categories. Here is a summary of the main points that came up (but where
not neccessarily agreed upon):
* Categories as they are have prooven rather useless and confusing (at
least in the german WP). Cross-Sections and transitive covers have been
agreed there to be the core features missing to create a functioning
structure of categories.
* Nikola Smolenski proposed a patch for linking to cross-sections of
categories using a syntax like [[:Category:Woman/German/Author]] to get
female german authors. That syntax would break some existing categories,
but changing the syntax is not a problem. The patch was discussed here
and dismissed as not practical (why exactly?).
* Cross-Sections of categories are only useful if we also have a
seach-page that allows for that feature. Possibly, this is even more
important than linking to cross-sections.
* Cross-Sections of categories are only useful if we can search over the
transitive closures (members of all subcategories, recursively). An
alternative would be to dispose of subcategories all together - but that
would require extensive redundant categorisation of articles.
* Besides linking to cross-sections, it would be useful and intuitive to
allow an article to be added directly to a cross-section. That is,
[[Category:Woman/German/Soccer player]] should put the articles into
[[Category:Woman]], [[Category:German]] and [[Category:Soccer player]].
* One problems with transitive closures are circularities in the
sub-category relation. As I can't think of any situation where such a
circularity would make sense, I think they sould be avoided alltogether.
Maybe the software could output an error when it is attempted to
create a circular dependency?
* Another way of avoiding circular dependencies and non-intutitive
results of transitive closures would be to allow categories to *belong*
to other categories, without being a *subcategorie*. IMHO subcategories
should be declared using a special syntax, like [[Supercategory:Foobar]].
* Another problem with transitive closures is efficiency: storing graphs
in a relational database is not trivial and generally ineficient. I
think it would be best to store all "implicite" memberships of an
article whenever the articles assignment to categories changes. But that
would mean that a *lot* of article-entries have to be updated when a
subcategory-relation is added or removed. I belive this point to be the
main issue with my proposal for advancing the concept of categories.
* The concept that is actually wanted (in the de:WP) is general
metadata, consiting (at least) of the dimensions Time, Space, Type (of
objet or article), Topic (field of study), and Keyword (loose
bibliographical collections). Using this dimensions together with
cross-sections and transitive closures would yield a truely powerful
structure. The existing concept of categories should be expanded to meet
those requirement - the distinction of the dimensions, however, do not
need to be hard coded. Naming-conventions would be sufficient.
Could you please give me your opinion about these points? I would be
especially interrested in knowing if anyone has a good ideas as to how
to implement transitive closures efficiently. I belive that to be the
core problem here... and it would be *extremely* helpful to have that
feature. If someone can give me a definite "forget about it" with a good
reason, I would be disappointed, but it would be OK too...
Thank you very much
Daniel
