On Tue, 10 Jan 2017 10:43:40 +0100 Matus UHLAR - fantomas wrote: > On 10.01.17 10:48, Emin Akbulut wrote: > >Recently we receive spam messages and SA cannot block them. > > why should it? They seem ham to is. > > >I've also checked the raw message at > >http://spamcheck.postmarkapp.com/ and score was very low either. > > > >I've trained the SA and it worked for a while but now it's useless. > > > > > >How can I prevent those spams? They look like poems > > >Message source: > >http://pastebin.com/nnN0jGw8 > > clear case of mistrained BAYES causing message being marked as ham. > you just have to re-train such spams as spam, it may take some time > (not very long) until it starts hitting properly.
The pastebin example was auto-learned as ham, it may be hard to counter this with manual training. bayes_auto_learn_threshold_nonspam should be set lower. If possible clear the Bayes database and train from manually sorted mail.
