Setting up a Bogofilter Filter

• organize email into spam (bad email) and ham (good email) and initialize bogofilter:
  • create a new mail folder to store your spam;
  • train bogofilter on what is spam and what is legitimate email -- this step only needs to be performed once and the more spam you have the better it will work:

    bogofilter -s < name_of_spam_folder

    bogofilter -n < name_of_ham_folder

    If you have not yet collected much spam, we recommend that you wait until you have at least 100 or so messages.

• create a .procmailrc file in your home directory. Your .procmailrc should look something like this unless you use one of mh/nmh/xmh/exmh to read your mail -- in which case your .procmailrc should look like this instead;
• create a .forward file with the following 1-line content:

   "|IFS=' ' && exec /local/bin/procmail -f- || exit 75 #your_login_name"
  

  where your_login_name is your username. Note that the double quotes and the pound sign in the line above are part of the content;
• make sure that both .procmailrc and .forward are readable to all;
• send yourself a test message and make sure you receive it (to verify that you have not inadvertently misconfigured things so badly that you cannot receive e-mail);
• keep an eye on your procmail log file (e.g., ~/Mail/procmail_log) for the next little while, to make sure that your mail is being sent to the correct destination.

Important Note: it's possible that bogofilter may sometimes flag legitimate mail as spam. From time to time, you should examine the mail that bogofilter saves in the caughtspam folder. You should never configure your mail filters to discard automatically mail detected as spam by bogofilter. See below for more details on what to do to deal with "false positives" and "false negatives".

Re-Training bogofilter: Inevitably, you'll get "false positives" and "false negatives". You should examine your caughtspam file from time to time to look for "false positives". "False negatives" will show up in your regular inbox. You can deal with both by periodically "re-training" bogofilter. Save the false negatives and positives into two separate files and do the training as follows from the unix command line:

• bogofilter -Sn < false-positive-file
• bogofilter -Ns < false-negative-file

The above helps to retrain bogofilter to better recognize what is legitimate email. In my experience, it will take a few weeks of diligently doing the above retraining to get good results.

Tuning Bogofilter: I found that I got better results if I tuned bogofilter to be a bit more aggressive than it would be by default. Of course, this required looking for "false positives" a bit more frequently at first. You can tune bogofilter by explicitly setting a few parameters for its operation in the file ~/.bogofilter.cf. The following contents for ~/.bogofilter.cf worked for me:

            db_cachesize=4
            robx=0.375672
            min_dev=0.375
            robs=0.1778
            spam_cutoff=0.500
            ham_cutoff=0.450