[Spambayes] SpamBayes Old and New (part deux)
dwallen at missinglink.com
Tue Jan 20 03:38:38 EST 2004
1) Yes, some Spam is being filtered. Some is also being routed to the
"Suspected" folder. Unlike the original version I used on my Win2K machine
shortly after the Info-World article, most spam ends up in my in-box these
2)Regarding how you train it: The version I first used that worked so well,
worked pretty much with the defaults. In fact, it worked so darn well that
I was afraid to change anything from the defaults. This is no longer the
case with the current version, or on WInXP, or my current configuration.
3)My inbox currently holds 75 valid emails (many not read) and about 7,000
spams (nearly all not read.) Since I posted the original email in this
list, I deleted about 5K spams, so the raining originally occurred on about
60 good and 10K bad. I have tried training manually, and have also set it
to "I haven't configured..." in the hopes that it would learn faster. In
other words, I go back and forth, with not much luck either way. It is not
clear to me which way I should leave it. It did not seem to matter
4) My false negatives' scores are quite low. My most recent undetected spam
shows up as "0.214000...". My most recent "Nigerian scam" spam scores "Spam
Score: 0% (3.88578e-016)" according to SpamBayes.
Date: Wed, 7 Jan 2004 18:31:24 +1300
From: "Tony Meyer" <tameyer at ihug.co.nz>
Subject: RE: [Spambayes] SpamBayes old and new
To: "'Dennis Allen'" <dwallen at missinglink.com>, <spambayes at python.org>
<1ED4ECF91CDED24C8D012BCF2B034F13046777EE at its-xchg4.massey.ac.nz>
Content-Type: text/plain; charset="us-ascii"
> Since then, my Win2K machine has died, and I have installed
> the latest SpamBayes versions on 2-3 new WinXP systems, one
> using Home, and 2 using Pro. With the newer versions of
> SpamBayes (or maybe it's due to Win XP), no matter what I do,
> most spam ends up in my Inbox.
You say "most", not "all", so I assume that some spam, at least, is being
correctly filtered? (In other words, the problem isn't that the program
isn't working, it's that the results aren't satisfactory.) If that's not
the case, let us know, and we can try and figure out what the problem is
(the log files help, a lot).
Basically SpamBayes works as well as you train it. How has the training
you've done on the newer machines differed from the old? Do you have a
roughly even balance of ham and spam training data? How many ham and spam
messages have you trained? If you look at the spam clues for false
negatives, can you see *why* they have been classified as ham?
More information about the Spambayes