Caretaker Antispam - Technical Details

The goal of a spammer is to increase the number of messages that are read by people. As there always will be a small % of people that click on a message, the only thing the spammer needs to do is to increase the delivery rate.

One of the methods to increase the delivery rate is to mislead the antispam filters, and spammers are constantly using new techniques to try to acomplish this.
Besides sending messages in normal text format or html, spammers also have used methods such as PDF Spam, Image Spam, Stitched Images, URL Spam, and Invisible Ink. Each method invokes a response from the developers of antispam filters.

Another method to increase the delivery rate is to prevent people to use an antispam filter. If end users deem an antispam filter as useles and deinstall it, all messages will arrive freely. One of the used methods to acomplish this is called Bayesian Poisoining. See the section below on training.

Researchers suggest that if a user should pay for every e-mail, this would be a possible solution against spam. Caretaker Antispam has implemented a method for paying for each e-mail that you send. But the method used by Caretaker Antispam - Hashcash - does not mean you are paying with money, but you are paying with computing (calculating) time.

Another important item in the reception of email is to prevent the user to get infected by malware that is send as an email attachment. For this Caretaker Antispam places suspicious attachments in the antispam folder.

Training

Experts claim that a valuable spam filter has to offer the user at the least the possibility to teach the Bayesian text filter the difference between spam and legitimate e-mail. The more e-mails the filter checks, the more it will learn and the better spam is averted - people say.

In Caretaker Antispam this function is left out on purpose:



Many people will have received a spam message just like the above mentioned. It consists of image spam and a whole lot of irrelevant, yet legitimate and platitudinous text. These texts are added by spammers in order to fool Bayesian text filters. If the user takes this spam message in order to train his spam filter, then only the irrelevant platitudinous text will be added to the Bayesian library (the text within the image is inaccurate). As a consequence e-mail messages with legitimate text will after a while be 'stamped' as spam (a so-called false positive).

Offering the user the possibility to train the spam filter generally sounds like a perfect solution. But fact is, that the spam filter will after a while move legitimate e-mails into the spam folder. That's why we regularly update the Bayesian library in Caretaker Antispam ourselves, in order to avoide such false positives.