[topicmapmail] Proposal: Extending the Role of Occurrence Elements

[Murray Altheim]

rmorikaw@gmu.edu wrote:
> All very good responses. Thank you.
>=20
> Here=92s the basic idea I have in mind.  Say you have a topic map that =
is shared within a given organization =96 e.g. governmental intelligence =
org.  The topic map, as meta data, exists in some sort of shared database=
 (relational, OO, native XML, =85).  In essence, the topic map is the bas=
is for your organization=92s semantic layer, or knowledge repository.
>=20
> An analyst is given the task to determine who and what events preceded =
911.  The analyst has initial =93validated=94 information as a starting p=
oint =96 e.g. names of suspect hijackers, but everything else is conjectu=
re.  He creates a new topic that reifies 911.  The analyst may suspect Us=
ama Bin Laden, but since it is not validated information, he does not wan=
t to modify the shared topic map by creating an association between 911 a=
nd UBL (i.e. to avoid other analysts from basing conclusions on his unval=
idated suspicion). =20

I think one way to look at the problem you describe and your proposed
solution to it is that you take the structures of Topic Maps too
seriously. By this I mean that your modeling is perhaps too tied into
the basic relationships within the Topic Map architecture and not
modeled at a level above. You need to create various association types
for the kinds of relationships you want to model, not use the actual
Topic Map graph relationships. For example, there could be an association=

called "isValidated", there could be even levels of validation, either
enumerated or fuzzy.

> So the analyst is faced with a dilemma that the topic map alone may not=
 sufficiently answer:
> 1.  need to avoid unvalidated information from being entered into a sha=
red topic map

As above, create a way to characterize the information in the Topic Map, =
don't try to keep
it out.

> 2.  no easy way to analyze a temporal sequence of events, and relate th=
em to one another directly

Not in the core Topic Map model itself. But you can create a set of event=
 topics and
associate each event with its timestamp.

> 3.  no easy way to weigh one event as being more relevant to the issue =
than another

Set up a series of topics establishing an enumerated list of relevancy, o=
r create a
topic whose occurrences are scalar values.

> 4.  and once the analysis is complete, no easy way to share analysis wi=
th analysts of=20
 > other organizations; especially if the other organization=92s semantic=
 layer is based upon another standard - - RDF for instance.

That's a different issue. A bit orthogonal. Transformation of content can=
 be done, but
it may be better to avoid having to in the first place.

> However, if the analyst can pull relevant occurrences (i.e. relevant to=
 his issue =96 i.e.  911), and connect them logically, he avoids many of =
these problems.  In other words, his =93occurrence=94 network, which resi=
des at the logic layer in a separate space for ongoing and unvalidated an=
alysis, provides a weighted (probabilistic) sequence of occurrences, that=
 can be shared (via XML and schemas), and contains links to the topic map=
 semantic layer (which in turn provides validated information).
>=20
> Upon validation of his analysis =93network=94, the shared topic map can=
 be modified with the new information. =20
>=20
> One last thing to note:  an analyst may have a =93gut feel=94 that two =
occurrences have a logical connection, but doesn=92t have enough evidence=
 to create a new association between parental topics.  He could, however,=
 connect these occurrences together in the =93network=94, so his =93gut f=
eel=94 is shared with others.  Once his gut feel is validated as fact, th=
en a new association can be created between topical parents (i.e. knowled=
ge discovery). =20

Make a topic called "hasGutFeel". Or give the association a low confidenc=
e rating. Or
a high one, depending on the weight you want for gut feelings (which ofte=
n are correct).

> Apologize for the lengthy explanation.  Does this make sense?  Is this =
old territory?=20

Makes sense, but you just need to model it *on* Topic Maps, not *in* them=
=2E

Murray

=2E......................................................................=
=2E...
Murray Altheim                         http://kmi.open.ac.uk/people/murra=
y/
Knowledge Media Institute
The Open University, Milton Keynes, Bucks, MK7 6AA, UK                   =
 .

    "The parties themselves, working with the Arab nations, have to find =
a
    way to co-operate to fight terror, without putting American forces in=

    an area where they will become targets."
                              -- White House Press Secretary Ari Fleische=
r.
                  http://news.bbc.co.uk/1/hi/world/middle_east/2989154.st=
m