[Spambayes] Important notify about your e-mail account.

[Tim Peters]

[Tim Stone, forwards and comments on a spam]

[Arvid Andersson]
> Now how would I mark your reply to that spam? Ham or spam?

Was it ham or spam to *you*?  That's how should you train it.

> It does contain a lot of spammy words, so I may pollute my database
> if I mark it as ham. On the other hand, I might want to receive
> comments about spam like this -it is a spam related mailing list
> after all.

Don't worry about polluting the database.  This is a "preponderance of
evidence" scheme, not a rule-based scheme, and a spammy word simply won't
become hammy unless it actually appears in more trained ham than spam.  If
it does, then continuing to *call* it a spammy word anyway belies your
actual judgment of the thing (and the system has learned enough to call your
bluff then <wink>).

> How do you experienced spam fighters mark mail such as this? Or
> those automatic on-vacation replies for that matter?

The real problem with "on vacation" kinds of things is that *I* change my
mind, and often, about whether they're ham or spam to me.  Then it just
doesn't work to train on piles of those as ham on Monday, but as spam on
Tuesday when I'm more pressed for time and so more resent dealing with
them -- garbage in, garbage out.  So I've found that the best thing to do
with Unsures like that is to delete them without training on them either
way.  I have tried training on them, but it's seemingly impossible for me to
judge them consistently day after day.

I also get plenty of Unsures where I have no real idea whether they're spam
or illiterate ham.  I imagine the latter are familiar to anyone on the
receiving end of a public "admin" or "help" mailing address.  It works best
for me to delete those untrained too.