Does "SA learn" feature really work?

Questions and answers about how to do stuff
Post Reply
maxkmv
Posts: 53
Joined: 28 Apr 2015 14:40

Does "SA learn" feature really work?

Post by maxkmv »

Or is it a waste of time?

Every spam email that I receive I add to Blacklist (IP address + domain name), I also diligently click the checkbox next to SA Learn and select "As Spam" - it produces a message: "SA Learn Learned tokens from 1 message(s) (1 message(s) examined)"

However next day I will receive exactly same or slightly modified spam, and the Score still says "Bayes probability 0 to 1%".

I have been doing this for 1 month (I add around 200-300 spam IPs and domains each day), but it doesn't look like its learning anything, meantime in EFA homepage it says "learned as spam" opposite those emails...

What am I doing wrong?

P.S. Note on many spam Bayes correctly detects spam probability as 50-90%, so Bayesian IS working, however I don't understand why it's not working for those types of messages that I add to blacklists - spam like "SEO", "webdesign in India", "business proposal", fake invoices, tax messages etc..
It just seems its only using some sort of pre-defined Bayesing rules, but not learning anything from my samples.
User avatar
pdwalker
Posts: 1553
Joined: 18 Mar 2015 09:16

Re: Does "SA learn" feature really work?

Post by pdwalker »

Someone correct me if I am wrong, but the blacklisting feature means that postfix won't accept mail from that incoming server at all.

Thus, your bayesian filter cannot be trained correctly because it is not receiving those messages.

Don't blacklist - let EFA accept it all and train your bayesian filters as you are doing and you should see that junk being marked as junk going forward.

You may have to increase your spamassassin weightings for your bayesian filters to be more effective.
Post Reply