Lists Home |
Date Index |
Watch what you eat for breakfast, this was a joke :-)
Now to address directly what you said about the semantic web, personally, I
convert Google's data into RDF and merge it with other RDF data in protégé.
IN other words, my own view on the semantic web is simply an assembly of
automatically and manually created frames (or call them triple sets if you
whish). Google is no worse, nor better than some of the manually created RDF
statements. I cannot say that Google is the semantic web vision but surely
said that it is a part of. Its only a meta data source having a certain
degree of credibility.
Now to defend the position of doctorow, I know that he said that simply
after observing how people try to manipulate the automatic acquisition of
meta data. Imagine now what would happen with manually created meta data....
Most likely, in that case, the semantic web would become associated to
something similar to spam: annoying with very lithe value.
To me, I can envision semantic webs limited to certain communities having a
good balance and check process. In other words: small islands of useful
meta-data in a see of crap. After the mosaic culture, the pop culture and
are now at the dawn of the 21st century facing the new era of crap culture
Now on the technical side:
A main problem when we try to aggregate disparate source of RDF frames is
that some manual processing has to occur.
a) not really useful and secure way to merge to identical triple with a
different value. For instance, two RDF elements about the same resource, one
stating <rating>good</rating> and <rating>poor</rating>. How to handle in
RDF two different points of views expressed as two different triple about
the same resource. The end result may be more complex unless you decide to
play Salomon and cut it for Good.
b) adding more triples is easier they are simply added to the existing ones.
However, again, I have to decide what makes the cut, is trustable, etc....
Overall, the trustable semantic web is a very intensive manual process
especially if your sources are automatic agents (fooled by people, by its
own developers by introducing bugs and bad algorithms), people (trying to
make a good impression to influence your action) and merging processes
(again with biases introduced by people and agent).
Other point of view.
It is very hard to distinguish links in RDF. Some are using the rdf:resource
element, some are using their own property name (ex: location), others are
introducing xlink:simple element as properties, etc... Very hard for a
generic processor to know what is a link pointing to an external resource.
So in addition with the other problems associated to the semantic web add to
this a meting pot of different vocabularies that a human being can process
but hardly done by a machine.
It is a lot easier with document. We human can decide if it is crap, what it
means and resolve the ambiguities. So the question without answers no: Is it
possible to envision a semantic web that can be processed with automated
agents? Maybe but with a very intelligent one. If you have one, please call
me, I will hire it, your RDF processor will as good as me, so I'll retire in
Bermuda and it will do the job, no more problems with outsourcing, you
processor will be cheaper :-)....
Didier PH Martin