Ed Kasky wrote:
At 01:29 PM Thursday, 4/3/2008, John Hardin wrote -=>
On Thu, 3 Apr 2008, Ed Kasky wrote:
X-Spam-Status: No, score=5.3 required=6.9 tests=BAYES_99,HTML_MESSAGE,
RDNS_DYNAMIC,SARE_OBFU_MILLIONS autolearn=no version=3.2.4
How did it hit SARE_OBFU_MILLIONS with a blank body?
I wish I had an answer for that one the same as why it didn't hit
BLANK_LINES_80_90...
Odds are the message isn't blank.. Have you got a copy of the raw
message before Eudora gets a hold of it?
Eudora will discard all but one of the text mime sections of a
multipart/alternative message prior to storing it in your mailbox. It
does this for space reasons. The basic reasoning is that if the MUA is
only going to ever render the text/html, there's no point in it keeping
the text/plain, so it gets truncated out.
The only way to get a hold of the complete message is to grab a copy
before eudora touches it. The copy stored by Eudora has been mangled.
That said, in response to your original post:
"Thanks in advance on this one. These things have been plaguing me for
some time and no matter how many I run through sa-learn, they never seem
to score above a 5... "
"X-Spam-Status: No, score=5.3 required=6.9 tests=BAYES_99,HTML_MESSAGE, "
Well, clearly that one scored above a 5. And with BAYES_99 already in
the mix, more sa-learn training won't raise the score. This message
already matches the highest bayes classification possible.
Perhaps you need to reconsider your threshold. If false negatives are a
big problem for you, raising it above 5.0 isn't a good idea. When you
raise the threshold, you're trading off fewer FPs, for more FNs. This
particular message clearly exemplifies that.
