KB 292

How do I get Spamprobe to be Trained?

Article:292
Created:2004-06-03 11:20:09
Categories:
  E-mail
  Spam

Question or Symptom

I just got Spamprobe turned on, but it says "Spamprobe is not Trained yet". How do I get it trained?

Resolution

This document describes how to train Spamprobe so that it can filter spam out of your mailbox.

Spamprobe is a program which inspects messages to determine if they are spam. It is similar to Spam Assassin in that respect, but there are some important differences:

1. Spamprobe gives each message a GOOD or SPAM rating, as opposed to a
   numeric score.
 
2. Spamprobe adjusts itself to *your* email.  It learns what you think is spam
   and what you think is good email, and classifies new messages accordingly.
 
3. Since Spamprobe adjusts itself to you, it needs "training".  That means you
   have to give it examples of good email so that it knows what your good
   emails typically look like.
 

On the one hand, this training requirement makes Spamprobe a little harder to start using than other spam filters. The good news is that learns quickly, and once it is trained up, it is VERY good at identifying spam. One user was receiving over 500 spam messages per day (even with Spam Assassin). After training to use Spamprobe, that number is down to less than 10 per day.

Here is how to start using Spamprobe:

1. The first step is to enable the Spamprobe rule on the SWCP mail filter web
  page (if you are reading this message that likely already done).
 
2. Spamprobe is now in "Enabled (Training)" mode.  You now need to start
  giving Spamprobe examples of your "good" email to train it.  The way to do
  this is by using the SWCP Webmail interface (TWIG).  Even if you normally
  use a different program to read email (Outlook, Netscape, Eudora, etc.),
  you need to use TWIG to classify some of your email messages *before* you
  download them to your usual mail program.
 
  If your mail program connects using POP, it should have an option to 
  "leave a copy of messages on the server".  While training Spamprobe, you 
  may wish to turn this on, so that the messages are available on the server 
  in webmail.  However, you *must* choose an option which does eventually 
  delete the messages (such as "Delete from server after 5 days").  If you 
  are unsure of how to do this, please call us and we can help you set that up. 
  If you are using IMAP, this is not an issue
 
  Then, go to https://roundcube.swcp.com/ and log in with your SWCP email
  username and password.  If your SWCP email address is "bob@swcp.com" then
  your username is "bob".
 
  In your inbox that, we want you to identify the ones that are good emails (NOT
  SPAM).  Click select each message which you are sure is a good non-spam message. 
  You can select multiple messages at once using the CTRL+left mouse click combination
  on Windows or Linux/UNIX, or with CMD key + click on Mac.
 
  Now click the "Good" button in the mail toolbar above your message list. This will copy 
  the messages to your learn_as_goodmail folder.
 
  If you doing this using another mail program over IMAP, It's important to COPY the 
  messages to the learn_as_goodmail folder instead of moving them. Copying ensures that
  a copy of the good message remains in your inbox.
 
  You do not need to train Spamprobe initially on spam.  We receive a ton of
  spam and take care of that part of the initial training for you.
 
3. After your learn_as_goodmail folder has enough messages in it (50+), the
  system will automatically grab them and use them to train Spamprobe for
  you.  At this time, they are DELETED from the learn_as_goodmail folder.
  That is another reason it is important to copy the messages there instead
  of moving them -- otherwise they will be deleted and you won't be able to
  access them again.
 
  At this stage, you'll receive a short email from the system to notify you
  that your Spamprobe setup has moved into "Enabled (Active)" mode.  Now all
  of your incoming email will be examined by Spamprobe and rated as either
  spam or non-spam.
 
4. At this point, you will need to start correcting any mistakes that
  Spamprobe makes.  To do this, go into Roundcube before you download 
  your messages with your usual mail program.  There are two types of mistakes
  that Spamprobe will make:
 
     A. Spam in your Inbox: any spam messages in your Inbox should be moved
        to the "learn_as_spam" folder. Select the bogus messages and click the 
        "Spam" button in Roundcubes mail toolbar.  Spamprobe will automatically 
        eat up these messages and use them to train itself to better recognize spam.
 
     B. Good mail mistaken as spam: Spamprobe will put any messages it thinks
        are spam into the "suspected_spam" folder. This is Spamprobe's actual 
        spam folder, so you don't have to do anything with junk that lands in it. 
        Most of what you see here will be spam. However, if you see any GOOD messages 
        here which should NOT have been categorized as spam, you need to copy them back 
        to your Inbox *and* copy them to the learn_as_goodmail folder, or click the 
        "Good" button in Roundcube.
 
        Any other messages in the suspected_spam folder can be deleted if you
        want to, or you can just leave them there.  The system will
        automatically delete them after 30 days.
 

When you first start this, it will seem a bit tedious. However, after just a few days, Spamprobe will start to become very smart about what is spam and what isn't. Some people will always want to check their suspected_spam folder to make sure no legitimate messages get through. Many people, after about 2 weeks of training, stop checking it altogether because the likelihood of finding a miscategorized message is so small.

Here is a summary of the steps:

1. Enable Spamprobe rule on the SWCP mail filter web page.
 
2. In Training mode: Copy "good" messages to the "learn_as_goodmail" folder.
 
3. After initial training with 50+ good mails, Spamprobe becomes active.
 
4. Correct any mistakes that Spamprobe makes:
  A. Spam in your Inbox: move to learn_as_spam folder
  B. Good mail in suspected_spam: copy to both Inbox and
     learn_as_goodmail folder
 
5. After 1-2 weeks: relax and enjoy a nearly spam-free mailbox

Related URLS

 https://kumo.swcp.com/spam/

Related KB articles

 KB 293 Ok, Spamprobe is Trained. Now what?