[topicmapmail] Re: Topic map design guidelines? (Jan Algermissen)

[Johannes Busse]

Jan Algermissen wrote:

> - You should also (from my experience) restrict yourself to

> =C2=A0 - using a given role only in associations of a single type (the
> =C2=A0 =C2=A0 combination of the roles is essentially what constitutes the
> =C2=A0 =C2=A0 type of relationship, so they should not be mixed across
> =C2=A0 =C2=A0 different types). This also enables you to select associati=
ons
> =C2=A0 =C2=A0 soly by role: "find all assocications where X plays supercl=
ass"
> =C2=A0 =C2=A0 as opposed to "find all assocications of type superclass-su=
bclass
> =C2=A0 =C2=A0 where X plays superclass".

hm, I was wrestling for a long time with this problem.

Jan: to model the assocs between relatives (son, mother, uncle etc.),
you would recommend a handful (up to a dozend!) of comparatively spezific
assocs, euch of which can be easily predicted once you know one of the roles
=2D- and vice versa? Then in fact the visualization of a topic map can
omit to tell you which roles take part within a given assoc (like the=20
omnigator does in the current version.)

But what about the following model: just take one single and comparatively
generic association (let it call "relatives"), and allow each of the
roles (like son, mother, uncle, sister etc.) to take part within this
association. The specific relationship between two persons is modelled
then not only in terms of associations, but also in terms of (tuples of)=20
roles. Then -- in contrast to the preliminary=20
model -- the visualization of a topic
map has necessarily to tell you not only the assoc, but also the roles
included (Hello Steve Pepper: we talked about this design principle some we=
eks=20
ago in Darmstadt. You considered to add=20
an new skin to the omnigator which makes the roles more explicit also in=20
binary assocs.)

more abstractly the main question is:
=2D Is it a good design to chain roles and assocs=20
  rather to such an extend, that your prediction will be sucessful?=20
=2D Or should on the contrary each of them (roles and assocs)=20
  represent some specific information, which might be traded=20
  rather independently in a model?

Well, one part of the answer is clear: it depends on the navigation
and visualization of an XTM, or the possibilities of your query
language. (I for my part work on a "role focusing" visualization of XTM;
for an unpublished proof of concept see
http://www.ews.uni-heidelberg.de/~busse/darmstadt04/bbb-fancy/home.html )

But the other part of the answer might be formulated in
terms of coding theory. here especially the problem applies, how
much redundancy within a model you want to have. And from this point
of view a concise model would not allow you to=20
"select associations soly by role" -- not at all!

Well, there is no conflict. But a question:
Do you agree that both modelling approaches=20
are "valid" according to the XTM specification?


herzliche Gr=C3=BC=C3=9Fe
Johannes Busse
=2D-
Dr. Johannes Busse,  Erziehungswiss. Seminar,  Univ.Heidelberg
Akademiestr.3, 69117 HD  |  Kellereigasse 3, 69 526 Ladenburg=20
tel 06221 / 54-7531      |  tel 06203 / 95 35 83
http://www.jbusse.de        mailto:busse@ews.uni-heidelberg.de