[Spambayes] Marking message flagged as spam as non spam

skip at pobox.com skip at pobox.com
Mon Apr 6 18:40:06 CEST 2009

    Keith> Essentially whats happened is that a user put one of his
    Keith> whitelisted ( I know that Spam Bayes doesn't recognise
    Keith> white/black lists, its just the easiest way of describing
    Keith> it)senders into the Junk mail folder by mistake. From that point
    Keith> on no matter what I've done I cannot make that particular sender
    Keith> be seen as non spam, apart from re-training from scratch. I've
    Keith> put all this senders mails into one of the 'ham' folders , and
    Keith> still its marked as spam, that is what I mean by having no
    Keith> effect. So essentially its seems that by just merely arranging
    Keith> mail into the appropriate folders isn't enough, I have to
    Keith> re-train Spam Bayes for it to take effect.  With regards to your
    Keith> other questions...

    Keith> Is it possible you have more than one mistake in your spam
    Keith> folder?  -Definitely not, ALL mails from the sender were moved
    Keith> into the ham folders.

    Keith> How many hams and spams have you trained on? 
    Keith> -Quite a few , around 350 spam mails, hams around 4500.

This is way out-of-balance.  Typically SpamBayes works best with roughly
equal numbers of ham and spam.

Can you post the clues SpamBayes generated for one of these misclassified
emails?  If you are 100% confident that all messages from the sender have
been trained as ham, there must be other spammy clues in the messages which
are causing it to be classified as spam.  Also, it looks like you are
posting via Thunderbird.  Are you using the pop3 proxy?  Is your colleague
using that or perhaps the Outlook plugin?  I don't have access to those GUI
tools here at work but I believe both allow you to retrain from scratch
using your current collection of ham and spam.

Skip Montanaro - skip at pobox.com - http://www.smontanaro.net/
        "XML sucks, dictionaries rock" - Dave Beazley

More information about the SpamBayes mailing list