Re: Spam spam spam spam spam

[Russ Allbery <rra@stanford.edu>]

Lars Magne Ingebrigtsen <larsi@gnus.org> writes:

> Now, about the white-list thing -- there should also be a black-list
> thing as well, I guess.  "Classify all mail from this From address as
> spam."  `C-u M-y', perhaps.

> What should the format of the black-and-white-lists be?  It could
> just be a one-address-per-line thing, but perhaps it would be nice to
> allow regexps?  Or perhaps a GLOB thing would be better?

> larsi@gnus.org
> larsi@gnus\.org
> .*@gnus\.org
> *@gnus.org

> I think it's easier for people to edit GLOBs than to edit the
> regexps, and I don't think people really need regexps here...

It would be great to somehow plug in the same logic that scoring has so
that we can use all of the same options that we can with score file
entries.  If you think glob would be particularly useful, it could be
added as a scoring type too.  :)

Incidentally, while this isn't a good heuristic for everyone, for those of
us who don't speak any Asian language and don't live in Asia, the
following is remarkably good.  All by itself, it catches something like
90% of all of my spam.  (Asian language spam has increased drastically in
the past year or so.)

 '(nnmail-split-abbrev-alist

[...]

         (cons 'content-spam
               (concat "big5\\|gb2312\\|ks_c_.*\\|shift_jis"
                       "\\|default_charset"))
         (cons 'subject-spam
               (concat ".*=\\?\\(big5\\|gb2312\\|ks_c_\\|shift_jis"
                       "\\|euc[-_]kr\\).*"
                       "\\|.*[¹²°¶÷¾].*"))

I then check content-spam against the Content-Type header and subject-spam
against the Subject header.  The last bit of subject-spam catches a lot of
spam with unencoded Asian languages in the subject header by looking for
characters that are fairly unlikely to be in the Subject for ISO 8859-1 or
-15 languages, but again isn't going to be applicable to everyone.

Not sure if something like this would be worth offering as an option or
example.  It's wholly inappropriate for people who correspond in Asian
languages, or in any language that uses those code points which may
include any UTF-8 encoding, but for those of us who only correspond in
English or western European languages it's staggeringly effective.

Obviously, this is the sort of check that you want to put after you
already split out your mailing list traffic, so as to not get false
positives on mail to public mailing lists from people who just want to
spell their name correctly.

-- 
Russ Allbery (rra@stanford.edu)             <http://www.eyrie.org/~eagle/>