> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On
> Behalf Of John D. Giorgis
[snip]
> >Then how, IYHO, should the analysis be carried out? What is your
> solution
> >for removing that what you perceive as a flaw?
>
> The answer of course would be real-time monitoring, where each post is
> filed as being either a "brand new" discussion, response, or possibly some
> category for "tangential exploration."
I'm working on this. But I'm re-doing the whole thing in Python while
contemplating whether to avoid infringing on the patent for what I
previously invented, which is somewhat in dispute. Although I mainly wanted
to learn Python, somewhere along the way, it dawned on me that I could
hardly infringe on other IP if I'm working in an entirely different
language.
All that aside, it may not be too long before I can show some more
interesting data. I have most of the building blocks written and debugged.
Now it's a matter of wiring them together with management UIs. Oh, and I've
completely redesigned the underlying data structures, which has been fun,
learning more about normalization and data design than I had really planned
on.
> I think that's the only way to get data that is more accurate than mere
> subject headers. Fortuantely, since I usually leave all previous subject
> headers in a subject line for Reference, that probably wouldn't be too
> hard.....
Not everyone does, but that sure helps. I'm doing some linguistic analysis
on the people whom the software judges as most significant, based on the
kind of connectivity I showed here a while ago, as well as things like
"excitation" -- the volume of postings and participants that follow a given
person's postings.
I'm curious if anyone here has any thoughts on potentially interesting
external events to look at in relationship to Brin-L discussions. When I
look at stock market discussions, I look for patterns relative to price,
volume and various derived indicators. For something like Brin-L, it's hard
to imagine anything so quantitative, but I could do comparisons to things
like New York Times headlines.
> Of course, this sort of thing could also be done on the historical data
> available, but I think that would be far more boring than trying to do it
> with current data....
Hey, that's my software you're calling boring! But seriously, the only way
to test and identify potentially interesting patterns is to run algorithms
against historical data.
Thoughts and suggestions are welcome, especially right now, since I'm
designing data structures that may be a real pain to modify later.
Nick
P.S. And I am chagrined to realize that I promised Jeroen some server space
more than a week ago and let it slip away. The machine is sort of
half-configured.
