[LINK] Chinese spam

Craig Sanders cas at taz.net.au
Fri Aug 25 12:05:58 AEST 2006


Brendan Scott wrote:
> Having received some spam in Kanji/Chinese characters this morning
> I wondered whether character based writing systems have better spam
> filters?  Presumably it would be harder to incorrectly "spell" a word

i have no idea whether it is harder to misspell a word in chinese or
not, but anti-spam systems don't work the way you think they do.

"creative mis-spellings" don't make it harder to detect spam, they make
it MUCH easier.

almost all modern anti-spam systems include bayesian probability
filtering. the more often it sees a word in non-spam (aka "ham")
messages, the lower spam probability it assigns that word. and the more
often it sees a word in a spam message, the higher spam probability it
assigns to that word.

so, words that are only ever seen in spam are assigned a very high spam
probability.

that (eventually) catches all of the annoyingly stupid mispellings of
common words.

it is also easier to manually create body/header check rules to block
specific spam phrases. for instance, i have a body checks rule in my
postfix config to block that "dr<at-symbol>gs" word that you included
in your message. i only saw your message because i saw the reject
line in my mail logs and went to the LINK archives to read it. i have
made hundreds of these rules over the years. they block a significant
percentage of spam.

(i have very effective spam filters. last week, for example, there were
17317 attempts to deliver mail to my system.  16351 or 94.42% were spam
and were rejected outright by postfix. of the 1044 messages that made
it through postfix, a further 78 or 7.47% were detected by spamassassin
and discarded. so 99.52% of all spam was rejected.  3 spams made it
through both postfix and spamassassin. these "Chinese Spam" messages
were the only false positives, 2 out of 17317 messages. without decent
spam filtering, email would be completely unusable for me).


in other words, spammers are making it easier to block their crap with
this stupid behaviour.


craig

-- 
craig sanders <cas at taz.net.au>           (part time cyborg)



More information about the Link mailing list