Imail's content filtering drives me nuts. Here's an example from an opt-in mailing regarding recipes. I know its opt-in because its my daughter's email account. I have three other clients who want the same content.
According to the laws of math, you can't alter the weight of a word. If you do, you break the math. It's like saying "How can I change 2 plus 2 so that it equals 10?"
Bayes Theorem (the math behind statistical filtering) just doesn't work by changing numbers (as opposed to a weighted spam system, where you can "tweak" the weights).
Most likely, the problem you are experiencing is that your daughter's legitimate mail is very different than yours, or that of other users. Since each person receives different types of legitimate mail (and different types of spam, too, which most people overlook), a statistical system needs to be trained for each recipient to work most effectively. It's one of the drawbacks of statistical filtering, as well as one of the benefits.
Whitelisting the sender doesn't work (and only seems to function about half the time, anyway).
Whitelisting should *always* work. I'm guessing the problem is that you are whitelisting in the wrong section (IMail v8's anti-spam -- which I assume you are using -- has a unique definition of "whitelist" that only applies to certain tests).
And whose idiot idea was it to weight 'walnuts' and 'teaspoon' so high?
A computer. Statistical filtering works on exact numbers. Either you received spams with "walnuts" and "teaspoon" in them, or whoever's stats you are using received spam with those words.
I've been thru the language file and it seems expressly set up to disallow food and cooking emails.
Most likely, whoever set up the stats received a lot of cooking spam. Perhaps they signed up for things at various cooking websites, and never agreed to receive E-mails from them. That's UCE they are getting, which most people consider to be spam. That's why statistical filtering needs to be trained for each recipient to work properly. Using generic stats can be fairly effective, but only if a large enough volume of spam and legitimate mail is used in the training. This sounds like a case where a large enough volume wasn't used (unless, of course, a significant portion of spam is about cooking).
Frustrated with a system that seems destined to dump out an unacceptable number of false positives.
For the benefit of others on this list, I'll behave, and not comment on this. <G>
11:12 09:25 SMTP(0D640002) word = mealtime, probability = 0.990000 11:12 09:25 SMTP(0D640002) word = bouillon, probability = 0.990000 11:12 09:25 SMTP(0D640002) word = teaspoon, probability = 0.990000 11:12 09:25 SMTP(0D640002) word = walnuts, probability = 0.990000 11:12 09:25 SMTP(0D640002) word = garnish, probability = 0.990000 11:12 09:25 SMTP(0D640002) word = pecans, probability = 0.990000 11:12 09:25 SMTP(0D640002) word = balsamic, probability = 0.990000 11:12 09:25 SMTP(0D640002) word = recipenotes, probability = 0.990000
The fact that they are all exactly .990000 says a lot -- first, that it was almost certainly the exact same spam that was received each time, which would in this case suggest that only one spam was received. The exact 99% factor would imply that those words never appeared in legitimate E-mails, and that the statistical filtering gives a maximum of 99% to adjust for having a small volume of spam to work with.
-Scott
---
Declude JunkMail: The advanced anti-spam solution for IMail mailservers.
Declude Virus: Catches known viruses and is the leader in mailserver vulnerability detection.
Find out what you've been missing: Ask about our free 30-day evaluation.
--- [This E-mail was scanned for viruses by Declude Virus (http://www.declude.com)]
To Unsubscribe: http://www.ipswitch.com/support/mailing-lists.html List Archive: http://www.mail-archive.com/imail_forum%40list.ipswitch.com/ Knowledge Base/FAQ: http://www.ipswitch.com/support/IMail/
