D2S uses Bayesian spam filtering to determine the probability that a message contains spam based on words and other characteristics found in the message. The list of characteristics is created by looking at words in the body of the message, its headers, HTML code, word pairs, phrases, and meta information. The filter then calculates the probability that a message is spam based on how it scores compared to previously analyzed spam and good e-mail.
For example, if D2S has 1,000,000 subscribers who all receive the same Spam mailing, the system learns from the characteristics of the e-mail and includes the information in the database. When I first heard the term Bayesian filter, I was clueless until I did a little research. If you would like a more detailed explanation of Bayesian filters, what they are, and how they work, visit http://email.about.com/cs/bayesianfilters/a/bayesian_filter_p.htm. This is a very easy to understand article by Heinz Tschabitscher that explains the good and the bad about Bayesian filters, spammers, and e-mail.
Setting up for D2S is easy. In practice, you will change your e-mail setting so that your mail will pass through the D2S server to be scanned and labeled as either spam or a legitimate message. Before a message is sent on to you, it will be marked as: [spam], which means that D2S is positive that it is spam; or [spam?], which means that it has many of the characteristics of spam but D2S isn't certain; or [???], which means that it has characteristics of both spam and good mail and D2S is unsure how to classify the message; [good?] means that D2S thinks the mail is good but is not certain. If a message is determined to be good, it will pass through to your Inbox unmarked.
I decided the best way to test this out was to use my husband's e-mail as the test subject. He gets very few good, or legitimate, e-mail messages and a lot of the worst spam has to offer. I thought it would be a good test for D2S and with so few messages it would be easy for me to do any training needed.
I changed his e-mail settings and created mail rules to send messages marked with the above classifiers to a folder called Death2Spam. I should also note that there was already a commercial spam filter installed on his computer, which did nothing more than mark a message with the word SpamAlert and dump it in the Inbox with the rest of the mail. Heck, I could have done that myself without paying money for that software. Besides, it was almost always wrong.
Much to my amazement, after about two weeks of filtering my husband's e-mail, D2S had an accuracy rate of 100%. All of the 97 messages received were correctly classified. Unfortunately this didn't provide any opportunity for me to do any training of the program, so I used my own e-mail as another test subject. I receive between 100 and 300 e-mails every day. They include personal messages, newsletters, survey invitations, weekly ads from places like home Depot and CompUSA, pay-to-read e-mails, and a huge number of the worst garbage e-mails out there. This would offer a real test for D2S.
I repeated the mail setup (detailed directions for setting up your e-mail can be found on the Death2Spam website at http://death2spam.net/docs/email-setup.html) as described above and created a mail rule to move the reclassified mail to a different folder. I clicked the button to retrieve my e-mail and watched as the Death2Spam folder started to fill up.
Copies of the messages will stay on the server for a limited amount of time to allow for the training process. Each day I would sign on to the D2S mail server to move any mis-classified e-mail to the correct folder (good or spam) so that it would learn from any mistakes it made. When I finished, I would usually delete all the messages from their server.
