Path: utzoo!mnetor!uunet!husc6!uwvax!oddjob!ncar!woods
From: woods@ncar.ucar.edu (Greg Woods)
Newsgroups: news.misc
Subject: Re: Take a sniff of Gnews
Message-ID: <40@ncar.ucar.edu>
Date: 24 Mar 88 02:20:49 GMT
References: <1501@looking.UUCP> <34@ncar.ucar.edu> <7928@agate.BERKELEY.EDU>
Reply-To: woods@handies.UCAR.EDU (Greg Woods)
Organization: Scientific Computing Division/NCAR, Boulder CO
Lines: 21

In article <7928@agate.BERKELEY.EDU> weemba@garnet.berkeley.edu (a small gnews-ance) writes:
>Thus, looking for "Guin?ess" in articles crossposted to 4 or more news-
>groups, or counting the number of times "abortion" or "fetus" shows up in
>a non talk.abortion article, etc. are not terribly difficult tasks.  There
>are 100s of these kludges that could be tried.
 
   Yes, there are. My question is: how useful is that information? I say:
not very. Someone might post a line in a comp.lang.c article like "I wrote
a really bad subroutine to do this; it was truly an abortion". 
Counting words is NOT GOOD ENOUGH to do classifications. You have to
determine if the word is really used in the context that would place it
into the classification for it (for example, the above article would
certainly not belong in talk.abortion). And THIS is what would require
"serious AI".
  As for the "fairy stories", I don't think it would be possible to keep
up-to-date with this sort of thing. Remember, lots of news admins are doing
that job as a 10th or 11th priority item. Maintaining a program as complex
as a GOOD keyword-based news system cannot, in my opinion, be done as a
10th priority item.

--Greg