GMail: an artificial neural network against spam


Google is strengthening its control system against spam mail using GMail its artificial intelligence. This is specifically an artificial neural network that detects spam more effectively while taking into account centers of interests of users to filter content more relevant.

In the sidelines of the presentation of his new tool Postmaster Tools which allows shippers to track the fate of emails they send to GMail , Google has formalized Thursday a significant evolution of its spam filter, probably already put in place for some time. “The spam filter now uses a neural network artificial ” and announced the team’s GMail .

The artificial neural network is a method of artificial intelligence (AI) Advanced which is now unanimity among large firms of the Web, which aims to approach the best operation of the brain human. It allows the machine to learn by itself to recognize, sort and analyze the elements that are submitted, which resemble others that have been presented and explained in the past.

Just as simply telling a child “it is a chair” for he then acknowledges all the other chairs with common elements that assimilates unconsciously, such as the four legs and the chair . The method is already impressive for the recognition of images or persons, but applies to many other areas.

Artificial intelligence closer to the habits

In GMail, the artificial neural network has two main functions:
  • Better detect spam attempting to impersonate normal emails, learning early on to distinguish new fraudulent emails characteristics reported by users (as is already the case for a long time thanks to spam-based Bayesian inference models, but with a new capacity for autonomous adaptation of the filter);
  • Personalize spam according to each user. Thus, even if the majority of users see no interest in receiving mails for Viagra or an enlargement of the penis , someone who would be interested in these topics would not see these automatically filed email messages in spam. More concretely, a weekly newsletter sent by a merchant may interest some of surfers and be considered spam by others. With its IA, GMail learns each user’s interest centers to customize filtering seamlessly.

The latter method of customization was already noticeable in GMail through the “important messages” that Google is able to bring out a mailbox overloaded, to allow each user not to miss emails it finds important to read . The ranking is done by learning the habits of Internet users, according to the content of the emails he has the habit of reading, it favors shippers, those he meets those he reads never, etc.