[Spambayes] My mail is being picked up as SPAM

Wed Dec 13 15:29:09 CET 2006

Bob may have a point regarding HTML in e-mails; perhaps I have fallen
for an urban myth. People I work with who want to send HTML e-mails have
often run afoul of various spam filters and believed that to be at least
in part because of the use of HTML. I don't know how rigorous their
determination was, or whether the situation at that time pertains today.
And I generally delete messages that are clearly spam without reading at
them (isn't that the point?), so I can't say whether the messages I
classify as spam are disproportionately HTML, or HTML messages are
disproportionately spam.

Bob, when you say "a high percentage of the spam I see is plain text,
with or without included JPEG images," are you saying you get non-HTML
messages with attached images, or HTML messages composed of undecorated
text and embedded images? I don't think there's any way to embed an
image in a text/plain message.

________________________________

From: spambayes-bounces at python.org [mailto:spambayes-bounces at python.org]
On Behalf Of Coe, Bob
Sent: Wednesday, December 13, 2006 8:43 AM
To: Mel Jensen; spambayes at python.org
Subject: Re: [Spambayes] My mail is being picked up as SPAM

We use Barracuda (a variant of Spam Assassin, IIRC) at work, and my wife
and I use Spambayes at home. Both work pretty well, and in neither case
have I seen any correlation between false positives and HTML. (Spambayes
doesn't give false positives; Barracuda does, but for other reasons.) IN
MY OPINION, the assertion that "the most important thing you can do" to
keep your messages from being labelled as spam is nonsense.

Just as there are some people who irrationally despise cell phones,
Windows, Unix, the Macintosh, etc., etc., there are those who see HTML
mail as the personification of the devil and rail against it at every
opportunity. (BTW, I don't include Jesse Pelton, who usually makes good
sense, among such extremists.) They're fighting a losing battle, of
course, but they do insist on fighting it; and their prejudices
sometimes seep into the common lore. You can try sticking to plain text
and see if helps, but I'd be surprised if it made a difference. FWIW, a
high percentage of the spam I see is plain text, with or without
included JPEG images. (This does NOT apply to "phishing" exploits; to
achieve a credible degree of deception, they pretty much have to be
HTML.) 

Bob 

________________________________

	From: spambayes-bounces+rcoe=cambridgema.gov at python.org
[mailto:spambayes-bounces+rcoe=cambridgema.gov at python.org] On Behalf Of
Jesse Pelton
	Sent: Monday, December 11, 2006 2:04 PM
	To: Mel Jensen; spambayes at python.org
	Subject: Re: [Spambayes] My mail is being picked up as SPAM

	The only thing you and your customers can do to avoid being
flagged as spam by Bayesian filters is to make sure your messages don't
look like what recipients classify as spam. That's a tall order, of
course; different people have different ideas of what constitutes spam,
and the language used in spam changes somewhat over time.

	Probably the most important thing you can do is to send only
plain text messages - no HTML. Unfortunately, that may be unpalatable to
your customers. If your system constructs HTML messages, I suppose you
could analyze the tags that are popular with spammers at any given time
and try to avoid using them, but you'd be running just as fast as you
could trying to stay where you are. (I'm not even sure that would work;
it depends on how filters tokenize HTML.)

	Other than that, you can assure your customers that some of
their messages will almost certainly be classified as spam. SpamBayes
does not automatically delete messages precisely because occasional
misclassification is inevitable. They might be able to reduce this by
asking themselves whether their messages sound like spam, but Bayesian
filters are often sharper than humans, so this might not work.

	At the risk of consorting with the devil, I'd suggest offering a
tuning service. Give your customers a way to run a trial message through
a number of spam filters and see how it scores, tweak it, and try again.
Maybe you could have consultants advise customers about how to adjust
their language. Maybe you could even make money at it. But part of the
reason I'm willing to make the suggestion where spammers can see it is
because I doubt it would work particularly well. Well-trained Bayesian
filters act astonishingly intelligently.

	For what it's worth, I would regard a SpamBayes spam score of 0
- 5% as "quite low." Almost everything with a score in the 40 - 60%
range meets my definition of spam. Of course, your filter may be
different.

________________________________

	From: spambayes-bounces at python.org
[mailto:spambayes-bounces at python.org] On Behalf Of Mel Jensen
	Sent: Monday, December 11, 2006 11:37 AM
	To: spambayes at python.org
	Subject: [Spambayes] My mail is being picked up as SPAM

	Hello,

	I'm a web developer, and my company provides an off the shelf
ecommerce solution, which in turn allows the webstore owners to
communicate to their customers via email.  We have recently changed
email systems to Mdaemon, and since then, some mails I receive from the
system are being flagged as follows:

	1.6 BAYES_50 BODY: Bayesian spam probability is 40 to 60%

	This in itself isn't too bad, as it's still quite low, and
varies from email to email, but I would like to issue some guidelines on
the site that certain things will trigger SPAM filters, and as we are
certainly not in the business of SPAM, and would remove anyone from our
system immediately should they be communicating to people unless they
had expressly opted-in, I'm wondering if you offer any guidelines for
companies such as ourselves.  

	My customers are getting annoyed that their customers, whom want
to receive information about special offers, are not always receiving
them.  We have introduced a whitelisting policy and guidelines, but I
want to have the best chance of delivering high quality, desired emails,
and I want to do it right.

	Thank you in advance,

	Mel

	T. 01784 419968

	F. 01784 419969

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.python.org/pipermail/spambayes/attachments/20061213/8beb5c1e/attachment-0001.html 