Frequently Asked Questions

What is SpamGAME ?

SpamGAME is a statistical anti spam filter based on the Gaussian Weighting algorithm for Text Categorization.
It is written in Python and should run on all platform having a Python interpreter. At the moment is has only been tested on a Linux box, I plan to test it on Win and Mac soon.
SpamGAME is realeased under the GNU General Public License.

How does it work ?

SpamGAME is meant to be used as a thin layer between the reception of the email from a user's email server and the collocation of it in the user's mailbox.
It works in a way similar to other spam filters:

• get an email message from standard input
• apply some kind of filters to the message and classify it as either spam or ham by appending a special header X-Spam: yes or X-Spam: no respectively
• returns the so modified message on standard output

Which methods are used to classify a message ?

SpamGAME uses different methods in sequence to classify a message:

• White List filter: a message whose sender is in white list is immediatly classified as ham (X-Spam: no)
• Black List filter: a message whose sender is in black list is immediatly classified as spam (X-Spam: yes)
• Forged Header check: a message whose headers are forged is immediatly classified as spam (X-Spam: yes)
• GA.M.E. categorization: statistical analysis of the content (plain text and/or html attachment) of the message and categorization based on the message history of the user (require training of the system). The message may be classified either as spam (X-Spam: yes) or as ham (X-Spam: no)

In what is it different from other spam filters ?

SpamGAME is the first spam filter based on the GA.M.E. (GAussian Mono-dimensional Environment) algorithm for Text Categorization.
GAME is a supervised text categorization method that in the context of SpamGAME tries to create a classifier of emails based on the hystory of the emails personally received and manually separated in spam and not spam (a.k.a. ham) by the user. With these 2 corpuses of emails the user can do the training phase in which the system GAME creates its internal configuration based on the patterns of the mail messages. When a new email arrives, the classifier calculates its similarity to each of the training corpuses (spam and ham) and assigns the message to the category with the highest similarity coefficient.
GAME is a statistical classifier method based on the same philosophy of the better known Bayesian methods: the user must traing the system with the messages he himself has received up to that point, in order to get an optimal classification of new messages.

Ok, but how much is it accurate ?

Based on personal benchmarks, SpamGAME should have an accuracy of around 95-98% in classifying new messages.