welcome > projects > 43

Fastnet SA : detect spams as soon as they arrive on servers

Mis à jour le 31/03/2016

Fastnet SA has made two improvements of spam tracking down to its software MailCleaner. The first concerns the location of text in an image, and the second, which uses artificial intelligence methods, concerns the detection of spam with a reliability of 100%.

MailCleaner is able to detect "hidden" text in a picture

Founded in 1994 and headquartered in St-Sulpice, Fastnet SA launched a CTI project in 2007 with two components to improve spam detection algorithms, thus ensuring the company of ten employees a step ahead of its competitors. The first component which searches text in images, was implemented in MailCleaner, the Fastnet’s flagship software. Meanwhile, the second component is still an experimental work in progress.

New analytical methods for data processing

The first part of the CTI project was quickly completed and implemented in the form of a software update. MailCleaner is therefore capable of detecting any "hidden" text in an image, a method commonly used during a certain period, by spams offering the sale of prescription drugs.

The second component of the project is more complex. The idea is to simulate the ability of the human eye to scan an email in a few seconds and determine if it is a spam or not, without even having to go through the entire text. Instead of reading each email as a whole text to locate key spam words, the algorithm is able to filter, in one operation, the entire email and analyze it as an image. Artificial intelligence methods were used for that before experiments could be carried out on the second module at Fastnet. Technological advances in this second component is important: the system is now capable of detecting spams with 100 % reliability.

Alliance identified good academic partners

Fastnet SA wished to apply for a CTI project, so they contacted Alliance, which then identified a good academic partner for their project: the research institute Idiap in Martigny. A first collaboration with Dr. Frank Crittin, then following his leaving a second collaboration ensued with Dr Yann Rodriguez; both went very well. Thanks to the CTI funding, Fastnet has developed algorithms for the development of two modules for the project.

Technological advances

The scientific success of this research seemed certain. The first module was immediately implemented. Since then, however, it has become obsolete, because spammers are changing their techniques quickly.

The module developed in the second part of the project is too big and too slow to be installed at the customers’ desktop system. Fastnet has used it to sort the so-called "false positives" and "false negatives "; that is recognizing good mails from being erroneously flagged as spams, or conversely, detecting spams that are not missed by the software. This artificial intelligence based module has enabled Fastnet to achieve significant progress against these false detections. It remains nevertheless too heavy and too slow to be used on the long term basis.

Although this project cannot be called a commercial success, it has however helped to make a step forward for this technology, which is considerable given the rapid evolution of spams.

Contact: alliance@alliance-tt.ch