Lists Home |
Date Index |
Going long and hazy. Apologies, it is getting late in the day.
From: Didier PH Martin [mailto:firstname.lastname@example.org]
>In fact, this research started when I discovered that the actual semantic
>web technologies as recommended by W3C don't help me aggregate different
>statement about a particular concept.
Aggregate or choose? In other words, given a term, there may be multiple
word senses (semioticians such as Sowa use the term, lattice of theories,
and backs that up with mathematical description, but not here). Systems
such as WordNet provide these senses.
>Actually, the main source of web document classification is Google.
An assertion, but fine.
>Let's consider that Google is offering a
>point of view or a view of the world. Some others like Teoma are proposing
>different point of view. If we have only one source of classification then
>no problems, we go back to the middle ages and get a single view of world.
>However, if, more than one proposition is made about a concept and if these
>sources a quite different, then, it would be useful to be able to merge
>in an intelligent way.
Why? They may indeed be different word senses and not resolvable to a
single meaning. When offered as choices, one picks one and uses it as
given, or alters it and publishes the new theory. Do you need to resolve
the difference or denote the difference? It seems if you are trying to
resolve all of them, you are back to the unified but medieval view, or
you have uncovered a cosmetic difference that makes no difference. I
can see the value in that but nothing that takes us to a new level of
>Just, for a moment, go to http://www.infospace.com/home/dog/index.htm or
>http://www.mama.com both are aggregating different point of views. We
>do not know how they come to the conclusion that a particular resource is
>more closely related to a concept than another. Without considering the
>social interaction or the why people link or interact in certain ways, how
>do I reconciliate different propositions. If people start to publish marked
>up information in the form of frames (ala RDF) and if people publish
>information in the form a text, how can I conciliate different propositions
>about a certain concept?
You can't. You have to test it in use and see if the test results match
the predictions. What kind of test that would be is an interesting
>If I am using the topic map method, I will collect a collection of
>about a particular topic. If I am using RDF I will aggregate propositions
>about a topic. Whatever the method I am using (topic map or RDF) I have to
>evaluate how related a resource (a proposition or a document) is to a
>concept/topic/class. The result needs both a notation (for presentation)
>an algorithm (for computation). This is operational, non philosophical and
>it does not depends on the linear or non linear nature of the web.
The operation isn't unless the algorithm behaves non-linearly. Again, I
suspect you are
denoting but not resolving.
>On the other hand, we have propositions
>based on formal and documented ontologies. For the latter, the problem is
>relate a particular statement to its ontology.
Which is the one-world view. So far so good.
>For example if I include the
>following statement in a web page and thus provide both text and data for
>two agents: a) a human being or automatic classification engine, b) a data
<rdf: description about="aparticularresource">
>The first problem is to get access to the ontology if there is one attached
>to this statement.
Yes. As in, does a URI act as a URL to the ontology for the namespace?
sort of problem. In that case, the resource identifies its own view of what
it means. So far, so good.
>I may have mine but as I said earlier my definition of a
>chemist may not be the same one as the author of this description.
Certainly. Without a meta-ontology, one that relates the definition
of chemist to the context of its use (Intention), you are stuck in
the classic Shannonesque choice: equally probable so fully entropic.
It is the hanged cat, or the Tic Tac Toe game with no winner.
>How to get access to the ontology behind an RDF description? Where is it
>located where in an RDF fragment? How do I get the link to fetch such
Access is easy is the URI resolves to the location of the ontology. If
not, you search. If you search, you are back to the Shannon dilemma of
having sufficient or insufficient means to choose.
>If the same document being categorized by one or several search engines
>provides an RDF description about this document and if the document also
>provides one, which algorithm can I use to resolve the conflicting
>descriptions about the same resource?
The problem is, if the descriptions conflict, you have to choose and
it the algorithm of choosing you have to define. A selector. So far
so good. Which definition meets your operational need most closely?
IOW, you have your own ontology and that is similar to the metaontology
I described above. Turtles all the way around, but one can't get past
the requirement of ontological commitment. One can use 'authority'
as a selector criteria; one can use 'favorite authors', and so on, but one
requires an ontology for those selections. There is a reason that
classical storytelling starts in media rex and storytellers learn
to write backstories.
>Is there one, none? Len if you think
>role playing can help; frankly I don't have a clue how. Can you publish the
>link pointing to this research so that at least I can analyze if their
>solution find its home for this type of problem. Because for the moment, I
>think I have a different ontology categorizing "role playing" than yours
Role playing is just a means of engagement to discover the source of
a conflict or to find a selector. I like it because given some set
of real world events, it is a remarkably accurate means to predict
an outcome. What if a way to create a selector is to
role play the conflict, observe it, and create the ontology that
controls the selection for the choices? What if by using simulations
of humans engaged in conflict perhaps with real humans (agent avatars
plus human directed avatars) one can do this online in real time?
I don't suggest that is the only way, but it is a way.
Why do topics in mail lists tend over time to not
reflect the actual contents of the particular emails? Think
about the debate some months back on the meaning of 'resource'
in web architecture. Even something that dominating as a
keyword has a very nebulous meaning. That is something
of what interests me because it demonstrates that no matter how
complete the ontology or how high the frequency of the term,
the humans will drift away from it and topical based tracking will get
On the other hand, will they as our exchanges have, orbit given topics?
The problemis that we can know what the attractors are, but can we predict
where any given email will land relative to one of them without a history
and when we have one, is it linear?