Spamato Reloaded Trust, Authentication and More in a Collaborative Spam Filter System

Spamato is a collaborative spam filter system implemented in Java. It is designed as a framework to support any number and kind of spam filters. The initial version features an URL Filter, which extracts URLs from incoming mail messages and calculates a fingerprint based on these URLs. This fingerprint is compared to a central database. If its fingerprint is known as spam, the mail message is classified as spam. Although Spamato is written in Java, a Microsoft Outlook Add-In has been written in Visual Basic.NET to use Spamato in everyday’s work. This thesis takes Spamato several steps further. First, a lot of improvements and extensions are introduced to the existing system, such as a new user interface in Outlook, a new user interface to configure filters, a new engine providing interesting statistics about the incoming mail of the user, multi-user capabilities as well as more consistent and robust communication engines to create XML messages. Second, a new spam filter is implemented and added to Spamato—the Razor/SpamNet Filter. Razor itself is also a collaborative spam filter network. It picks random text passages from mail messages and calculates a fingerprint using them. This fingerprint is used to report spam messages to the Razor network and to check if a message is known as spam. Using the Razor/SpamNet Filter, Spamato is able to check incoming mail using the Razor network in addition to the URL Filter network. Third, the Automatic Authentication and Authorization System (AAAS) is presented. AAAS uses email adresses and key sets instead of username and password to authenticate users. Therefore, no user action is needed in order to register a new account or to reregister an existing account after reinstalling. It is even possible to use the same account on any number of systems at the same time. Fourth, a new trust system is introduced. Trust systems are designed to protect collaborative systems against malicious users. Most of today’s trust systems, however, use a kind of AIMD (additive increase, multiplicative decrease) approach and therefore fail if no global classification of a mail message can be achieved. This thesis presents Trooth, an advanced trust system which is able to handle even these messages correctly by using a new approach for cases where no global decision is possible.