[pptp-server] spam spam spam spam
Andrew Kohlsmith
akohlsmith at benshaw.com
Tue Apr 16 10:56:07 CDT 2002
> Nice thought in theory, but in practice it sucks. The rate of false
> positives will be very high. Rather than bore everyone with the details,
> there is a (somewhat long-winded) explaination about SpamAssassin here:
> http://www.talkbiz.com/assassin.html
As someone who uses SpamAssassin for a 4000-user dial-up ISP I can tell you
that my rate of false positives is *very* low (on the order of less than one
false positive per 20-25 days). That URL you linked to is sensational crap
at best, and outright fear-mongering and misleading at best. He clearly does
not (nor does he appear to care to) understand how SpamAssassin works. My
users regularly get the stereotypical "FW: FW :FW :FW F:W:::FW::FW::WW READ
THIS!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!" messages and they are *not* tagged.
Block capital messages? Not tagged. Financial newsletters filled with words
like "Investment" and "buy" and "call" are not tagged. Steamy emails between
lovers are not tagged. Oddly enough, real estate agents send very
spammy-looking emails and they are not tagged. True opt-in mailing lists for
marketting are not tagged. How much better do you want to get?
We receive about 5000-7500 spam email messages a day and when I was first
setting this system up I went through the spamfile by hand looking for false
positives so I could fine-tune the tests. I no longer have to do so. And on
a list such as pptp-server, I can almost guarantee the false positives to be
zero due to the non-newbie, non-sales nature of the list content.
I suggested SpamAssassin specifically because of how it worked. Hundreds of
individual tests with scores based on a genetic algorithm and corpuses of
spam and not spam to run the algorithm against. Obviously I cannot (nor do I
wish to) force the issue; I suggested something which works for me (and
consequently for many, many others as well).
Regards,
Andrew
More information about the pptp-server
mailing list