Michael Grant schrieb:
The main drawback I see using this at a server level is that crm114 is
best trained by telling it that it got something wrong as opposed to
feeding in a lot of ham and telling it it's ham or spam and telling it
it's spam. I guess you would say that this child learns best by being
punished!
As a user you can safely ignore the train-on-error strategy.
If you just feed a lot of spam and ham into the crm114-SA-plugin (or
directly into mailreaver.crm/mailtrainer.crm) then it will only train
those mails which are not already classified correctly.
crm114 when it gets it wrong. This is because the result of crm114
would be burried inside a spamassassin tag as only one metric of
spaminess.
That's certainly an issue.
As an advanced user you probably want to set up a filter on hits for
CRM114_UNSURE, CRM114_PROB_SPAM, and CRM114_PROB_GOOD -- these indicate
that crm114 could not classify for certain and the mail should be trained.
--
Martin
