Lorenzo Lucioni wrote: > I configured SpamAssasin with a 2.0 points as threshold because many > spams came with a score lower than 3.0.
The spam scoring 3.0 is your problem. Fix that. Do NOT drive your score threshold through the floor to try to catch spam, as all you're going to wind up doing is creating more problems such as this false positive. Whenever you lower the threshold, your false negative rate goes down , but your false positive rate goes up significantly. Take a look at the data from STATISTICS-set3.txt: # SUMMARY for threshold 5.0: # Correctly non-spam: 29443 99.97% # Correctly spam: 27220 97.53% # False positives: 9 0.03% # False negatives: 688 2.47% # SUMMARY for threshold 2.0: # Correctly non-spam: 29277 99.41% # Correctly spam: 27696 99.24% # False positives: 175 0.59% # False negatives: 212 0.76% By dropping your score from 5.0 you've theoretically reduced your FN rate from 2.47% to 0.76%, a factor of 3 fewer spam messages will get by SA. However, you've also increased your False positive rate from 0.03% to 0.59%, 19.6 times as many messages will get tagged when they should not be. Thus, you're getting the exact behavior that was expected... You've jacked your threshold down low enough you're getting a significantly increased rate of false positives. > I would suggest to the person who send this newsletter to apply a > correction to his emails to avoid the: > "1.2 MIME_HTML_ONLY BODY: Message only has text/html MIME parts" > but I don't understand what it does mean. Can you help me and suggest my > how to modify emails to avoid the matching with this rule? That rule means just what it says. It got a mime encoded message, which only contained text/html parts. No text/plain or other types of mime sections were present in the message. This is pretty common for HTML newsletters as well as spam, which is why this rule only scores 1.2. Normally this rule hitting nonspam email would be no problem, as the message in your example only scored 2.0, which is less than half the required score for a spam tag in
