On Tuesday, Oct 14, 2003, at 20:42 Europe/Rome, Conal Tuohy wrote:
Stefano wrote:
"automatic harvesting" scares the crap out of me, Conal.
This is conceptually no different to harvesting JavaDoc tags from Java source.
good point.
I agree that there must be some kind of automatism going on, but the topic creation is a human task and programs would do a terrible job at doing this.
The example I gave assumed precisely that a human editor had written a namespace topic; the harvester was simply linking a document (which mentioned that namespace) to that existing topic. So this is automatic creation of associations or links, rather than topics.
yes, but this is simply spreading the issue of topic creation all over the place, you are not making it any easier (IMHO)
But topics can also be safely created automatically in some cases: where
good structured metadata exists we can confidently base topics on it. e.g.
topics can usefully be automatically harvested from Java classes that
implement particular interfaces (generators, transformers, etc).
True, but again, I don't see the point. I'm sure that if we make the editing interface to our doc system people will find it much easier to just make a list of components and update them as we go (expecially since they are not so many).
but anyway, we decided to do a first step with handwritten linkmaps. we can move incrementally from there on.
Yes that's true.
What I particularly like about TM is that they invert the usual relationship
of resources to metadata - in a TM the topics are central and the resources
are attached to them. So the key activity is to identify the high-level
topics (the ontology) and then build a harvester to link your resources to
the topics in the ontology. This linking can be done by recognising patterns
in the resources (e.g. a reference to a namespace), or, better, by
recognising explicit metadata (e.g. JavaDoc).
Very true.
-- Stefano.
