[Spambayes] Trial use experience

[Skip Montanaro]

    Dick> It now seems that my Inbox is 100% free of SPAM -- hooray!

Good to hear.

    Dick> I still must carefully scan my JunkMail folder for false
    Dick> positives.  I still find good email classified as SPAM even after
    Dick> I have recovered previous messages from the same sender.  

Yes, if you have valid correspondents who tend send you "spammy" mail, it
may take a little effort to get their messages classified properly.  Make
sure you train them as ham.

    Dick> Only about half of these wind up in my MaybeSpam folder, while the
    Dick> rest go directly to JunkMail.

I strongly suggest you train on everything which winds up in your MaybeSpam
folder.  Don't simply delete them.  Also, add a Spam column to the message
header display in both your MaybeSpam and Spam folders and sort the display
by that column.  Perhaps you need to move your spam cutoff closer to 90%.  I
think the default is 80%.  The spam scores in that column should help you
identify a better cutoff for your current situation.

    Dick> SpamBayes needs to have a real "White List" to pass mail from
    Dick> known sources directly to the Inbox.

SpamAtBay (the commercialized offshoot of Spambayes) has a white list
feature, but Sean True said recently it's not all that helpful.  In
addition, most spam and virus messages come with forged From addresses.
Viruses especially are likely to come with From address of someone you know.

In short, who the message is from is just another clue.  Spambayes generates
synthetic tokens for email addresses.  Your correspondents' email addresses
will become good ham indicators over time.

Skip