Quote:
Originally Posted by janusz
My impression is that the new system is much more aggressive than the old one: I'm getting numerous false positives.
In fairness, the wrongly marked messages are newsletters, which are notoriously difficult to correctly classify as spam/ham.
|
Reporting them as misclassified will help as there are per user, per provider and global feedback on this.
Quote:
Originally Posted by gecko
I just read this comment on your blog post:
https://blog.runbox.com/2017/02/new-...comment-175984
I think Josh raised a very important question there.
From your above statement, I understand that no data is sent to anyone outside Runbox.
However, some detailed description of how the spam filtering works and what the difference between "Train using reduced email details?" yes/no is would be much appreciated.
Many thanks and BR,
gecko |
If you train using reduced email details, then the reporting will be less effective, but it will still have an effect.
The reduced details options means that the subject and body is not reported back to Cloudmark, just a hash of the email. They may find a match of this hash or they might may not. If they dont have a match, then it is difficult to adjust anything.
Cloudmark also attach a trust value to each user reporting spam and ham and if what you report is "correct", then your reporting will have a larger effect.
Kim