Uses of STRN scoring--extracts from a letter.

> I like the idea of ranking output according to some score but the
> scoring rules seem a little thin. What about body-text searching for
> instance? Even this is just another boolean string-matching technique
> though.

	Body text searching may be an option later, but it is
(comparatively) very slow, and doesn't seem to yeild much benefit.
Speed is a major concern.  One problem is that one can't
(realistically) cache all the article bodies which leads to slower
rescoring times.  (Rescoring is necessary since the rules may be changed
while running.)

[...]

	The main use of my scoring code is to separate good authors
from bad authors, where good and bad refer to my interest in their
writing topics and styles.  I also use it to reduce crosspostings to
groups I have little or no interest in (such as talk.abortion and
alt.flame).  Finally, I occasionally use it when a particularly
interesting subject pops up in a newsgroup.

	The "Keywords:" line is only useful in a moderated group where
the moderator adds the keywords.  Most of the time that a Keywords:
line is used it is either a summary or some cute thing like "No one
ever reads these anyway."  In the clarinet hierarchy (UPI news wire
feed) the keywords could be very useful--I have tried it on a small
scale in the biz.clarinet.sample newsgrop.  The "Summary:" line is
useful to display (when present), but rarely useful for simple text
matching.

	The author of the article appears to be the best easy-to-find
indicator of its quality.  For instance, even though I am interested
in Intel's new chip the "Pentium", I am not really interested in just
anyone's ramblings about it.  However, I would be interested in
postings by people at "intel.com" about new chips.  So my "comp.arch"
scorefile has a +10 to articles from intel.com, and a +2 to articles
with "Pentium" in the subject.  In the comp.sys.amiga subhierarchy I
add 25 to all postings from commodore.com.  I also have a global
list where I put people who have consistently written very good
articles.  People like Eugene Miya, Russell Turpin, Henry Spencer, and
a handful of others are there.  (I also have a bonus for people
posting from my own site so I know what happens locally.)  There are
many, many uses of author-based scoring on USENET, more than I can
easily repeat here.  (....nasa.gov in sci.space, medical doctors on
sci.med, physicists in sci.physics, etc...)

	On the other hand, negative scores give me reasonably fine
control over what I want to read based on my current mood.  There are
some people that I have "banned from the net" in my view of it--they
are generally people who consistently write inappropriate or rude
articles.  Once I give them a global negative score low enough to
pass below the global KILLthreshold, I never see them unless they
change their from: line.  There are also a lot of people, however, who
aren't really *terrible* writers, but who rarely write things I'm
interested in.  For instance is people who seem more interested in
debating than discussion--they get a more moderate negative score, and
are left untouched by default.  If I'm busy, however, or just don't
feel like reading debates today, I can easily junk lower-scoring
articles.  There are several groups that I wouldn't otherwise have
time for, but now I just read the high-scoring articles
(rec.arts.books and talk.politics.misc are good examples).

	One thing has become clear in the past few months of using
scorefiles--the vast majority of very good articles are written by a
very small number of people.  An overwhelming majority are ones that I
am indifferent to, with a small number (~10%) of annoying posters and
very few which are not worth consideration.

	I do have a vague notion of categories in a different way (I
think): "interest groups". For instance, I have a scorefile called "politics"
where I put scores relating to people's views on politics.  While a
good author is generally worth reading anywhere s/he posts,
intermediate ones can be harder.  I have found a few people, for
instance, who are very good in one field (usually scientific), but
fairly poor writers in another (like politics).  With interest files I
can have context sensitive rules.  Currently, the only allowed context
is the group one is reading, but I hope to expand that to have
multiple kinds of contexts (for instance, one might have different
categories of "rubbish" depending on how much time they have to read news.)

> Leaving aside the way messages are matched, my feeling is that leaving
> it up to the user to manually specify rules is not the way to go.

	Definitely not *the* way, but *a* way nonetheless. :-)  What I
have now is a fairly simple-to-implement method that could definitely
use improvement.  I don't expect novices to use STRN for awhile--then
again anyone who can really *use* RN/TRN effectively is probably not a
novice :-).  I have a vague vision of an interactive database which is
easily updated and changed, but that will be some time in the future
(unless I find some code out there or someone writes it for me :-).

[...]

	This is where rating services would be very useful.  The
simplest service might consist of occasionally releasing scorefiles
which contain ratings of authors and threads.  More complex ones might
attempt to rate each article or add keywords to each article from a
standard list.  I hope that at least a few people will volunteer their
opinions of who writes good articles, or maybe even "moderate" a group
by sending rating-messages.  (If moderation is on a simple rating
scale, it isn't hard to do: I did a test on rec.humor by making macros
for function keys f6-f10.  Each key would give the article a different
rating (-10, -5, 0, 5, 10) and save it to the end of a file in the
scorefile format.  All that is necessary now is some method of
distribution.

> How far along is the stuff about propagating editorial changes with
> hypertext links and this idea of rating services?

	Well, aside from the small expiriment with rec.humor, not far
at all.  Then again, I like to call my work the "A News"
approach--write something which meets a need and other people will
improve it for you.  As I (unclearly) mentioned in the alpha notice, I
don't expect STRN to be able to handle the really advanced stuff like
changing articles or virtual newsgroups--another newsreader will
probably need to be written.

[7/28/93: Well, maybe I was wrong about the virtual newsgroups. ;-]
