Browse Prior Art Database

Method and System for Detecting and Filtering Spam

IP.com Disclosure Number: IPCOM000249540D
Publication Date: 2017-Mar-03
Document File: 5 page(s) / 135K

Publishing Venue

The IP.com Prior Art Database

Abstract

A method and system is disclosed that utilizes Generative Adversarial Networks (GAN) for training both a defender agent and an agent that generates emails that appear similar to normal emails in order to classify and detect different forms of spam. Thus, the method and system utilizes the GAN to better discriminate normal emails from spam and simulates potential future forms of spam for effective spam detection and filtering.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

1

Method and System for Detecting and Filtering Spam

Email is one of the most important form of communication and with the widespread use of emails, fraudulent practices such as, but not limited to, spamming and phishing have become inevitable. Spammers try to flood inboxes with unsolicited information such as, but not limited to, marketing information, false links and advertisements (ads). On the other hand, phishing aims at tricking an email recipient into believing that the email comes from a legitimate source in order to steal passwords and to infect the recipients’ system.

In order to combat such fraudulent practices, different solutions are known for filtering spam from legitimate email. One common approach to spam filtering is to use machine learning to learn and classify emails into two categories namely spam and legitimate email. By using known datasets of both legitimate email and spam, these techniques detect and exploit patterns for filtering the spam. Moreover, these detection mechanisms exploit Artificial Intelligence (AI) techniques such as, but not limited to, Naïve Bayes techniques and complex Neural Networks for spam detection.

However, the quality of the spam detection schemes mentioned above is directly dependent on the quality of the collected set of spam emails. Moreover, attackers come up with different spamming techniques and strategies, thereby generating new types of spam emails. Also, the attackers craft emails that are sufficiently different from the ones contained in a current training set and the crafted emails are similar enough to normal emails in order to escape the filter. Therefore, these techniques always require an up to date training set and therefore spam detection with such techniques may fail if spammers come up with new and different schemes for sending spam emails.

Thus, there exists a need for a method and system for effectively filtering spam emails from legitimate emails.

Disclosed is a method and system that utilizes Generative Adversarial Networks (GAN) for training agents in order to classify and detect different forms of spam.

The training process is modeled as an adversarial game between a generator network and a discriminative Network, wherein the generator network generates emails that appear to be as close as possible to normal emails in order to train the discriminative network. The discriminative network, then, functions as a spam filter to recognize spam from normal/legitimate emails.

2

Thus, the method and system utilizes the GAN to better discriminate normal emails from spam and simulates potential future forms of spam for effective spam detection and filtering.

FIG. 1 illustrates an architecture of the method and system in accordance with an embodiment.

Figure 1

As illustrated in FIG. 1, the method and system includes two steps: the learning step and the operational step. The goal of the learning step is to train a spam filter while the operational step filters real/legitimate emails...