Hi Bart,
A few quick questions/points:
- Do you have a default TB (Training Buffer) set or do you set this
per-user? TB helps reduce the number of FP (False Positives).
- Beyond just the final result (Spam,Innocent), what else have you
learned about the FPs? When you review the dspam.debug log are the
tokens it is finding and using to classify the message as spam, legit?
- Have you tried not including a domain in the merged group to see if it
has the same effect? I think you can set "ignoregroups" in the user
Prefs if you want to just have a specific user ignore the group. of
course, if you do this, monitor your dspam.debug and make sure the
merged group is not referenced in processing transactions for this user.
Sean
Bart Verwilst wrote:
Hi guys,
I've been using dspam with pretty great results. However, i still come across
one strange thing when adding new users.
All my users are in the form of "@domainname.com" . They all have a "fallbackDomain on"
setting, meaning they are "catchalls" for every user in the domain. I have one merged group, that
gets fed with 1/3 ham and 2/3 spam.
Now, when i create a new user, the first mails that are sent are almost all
quarantined, even when they're obvious hams. When relearning a few of them as
ham, the false positives go away. The problem is, we will soon migrate over
10.000 domains to the dspam solution. We cannot have it that almost all mails
are quarantined the first day until they learn them as ham.. Any idea what is
causing this? And how to resolve this? :)
Thanks a lot guys!
Kind regards,
Bart Verwilst
