Spam spam spam spam...

[wortgames]

I have decided to wage war on spam, despite my long-held belief that trying to do so results in more stress and frustration than just deleting it.

I am absolutely baffled by Pocomail's spam features - I am not a programmer by any stretch so I am relying on existing features to do the job for me.

I have got a long list of filters, most of which seem to get ignored, and I am obviously failing to teach Mr Bayes the difference between "Hey you - buy viagra" and "Dear Mr Wortgames we would like to confirm the following order with your company".

Over time I have adjusted many of Pocomail's spam settings, obviously in the wrong direction.

Can anybody tell me what this means:

X-Poco-Score-Detail: -100 [%BAYES%=P=0;T=90;BIAS=+20] (%bayes% P=0;T=90;Bias=+20)
X-Poco-Score-Detail: +7 [X-POCO-SCORE=+]AND[TO=%EMAIL%]AND[TO=%REPLYTO%]AND[CC=%EMAIL%]AND[CC=%REPLYTO%] (X-Poco-Score +, and To %email%, and To %replyto%, and Cc %email%, and Cc %replyto%)
X-Poco-Score-Detail: +2 [FROM=%ADDRESSBOOKS%] (From %addressbooks%)
X-Poco-Score-Detail: +100 [SUBJECT=%JUNKSUBJECTS%] (Subject %junksubjects%)
X-Poco-Scored: +9

This was a spam message, absolutely riddled with 'banned' words, a specifically 'banned' subject line, the sender is NOT in my address book and I have tried to teach the Bayesian filter it is spam several times, declassified it as good several times, and classified it as junk several times, and it still manages to sit there in my inbox defiantly.

Is there a limit to how many words / characters the 'banned subject' file can contain?

Is there a generally-accepted 'best practise' setup I should be using?

Does each 'filter run' leave its results in the message, to confuse further filter runs?

It seems the more I try to teach Pocomail about spam the less effective it becomes.

(FWIW I'm using 3.4.0.2130 with XP SP2).

Any tips would be gratefully received!

~ Moved from PocoMail to Junk Mail Filtering ... Eric ~

[Michael]

The meanings of the various Poco-Score-Detail lines is:

Code: Select all

X-Poco-Score-Detail: -100 [%BAYES%=P=0;T=90;BIAS=+20] (%bayes% P=0;T=90;Bias=+20)

The Bayesian filters thought this was not a junk mail message, 100 points deducted from the overall score.

Code: Select all

X-Poco-Score-Detail: +7 [X-POCO-SCORE=+]AND[TO=%EMAIL%]AND[TO=%REPLYTO%]AND[CC=%EMAIL%]AND[CC=%REPLYTO%] (X-Poco-Score +, and To %email%, and To %replyto%, and Cc %email%, and Cc %replyto%)

Not too sure what this test is doing, it appears to be making some comparison of the To, CC and ReplyTo lines. Probably looking for your email address (this is just a guess). The test determines that this aspect makes the message appear to be junk so 7 points are added to the overall score.

Code: Select all

X-Poco-Score-Detail: +2 [FROM=%ADDRESSBOOKS%] (From %addressbooks%)

The message is from an unknown sender, add 2 points to the overall score.

Code: Select all

X-Poco-Score-Detail: +100 [SUBJECT=%JUNKSUBJECTS%] (Subject %junksubjects%)

One or more words in the subject is in your junk subjects file, add 100 points to the overall score.

Code: Select all

X-Poco-Scored: +9

This is the resulting score, it will not result in the message being classified as junk. Your exact threshold is a setting in the Junk Mail filter options.

[wortgames]

Thanks Michael.

I think I've found one of the causes of my problems - according to one of the help files, moving messages from the Junk Mail folder automatically declassifies it as junk. I keep all my junk mail in a separate 'spam' folder for use in further learning and testing - unaware that every time I moved it I was inadvertently teaching the bayesian filter that it was not junk

Can you tell me - does 'copying' a message from the junk mail box also teach the bayesian filter? If so I can copy it to the junk mailbox and then delete the original?

I would still be interested to learn if there was a widely accepted 'best' way to set up the filters, they are very complex for a newbie to understand.

[rosscassell]

Thanks Michael.

I would still be interested to learn if there was a widely accepted 'best' way to set up the filters, they are very complex for a newbie to understand.

An effective way is to run POPfile, its an email classifier that runs as a proxy between the email server and the client. It uses bayes as well but its much more effective than the bayes implemtations that come with many email clients.

Its free and its here:

http://popfile.sourceforge.net