High Speed and Reliable Anti-Spam Filter

Unsolicited Commercial Email, also known as spam, has grown exponentially in the past few years and it is the biggest problem facing email applications. In this paper, we present a new anti-spam filtering technique based on Bayesian approach, where the training phase of the filter is enhanced by having the filter parse the header, the subject, and the body of the email message separately and independently. The filter also provides an auto generated, that we call it "Virtual Blacklist", which is used to speed up and improve the filtering capabilities. Moreover,, in the proposed filter, the frequency of the token is considered in terms of the size of the message. The filter is implemented at the application layer using Java for flexibility, speed and platform independency, then applied to various available corpuses and its performance is compared to the commonly used filters. Experimental results show that the proposed filter can achieve over 98.7% filtering accuracy at speed more than 10 times faster than existing filters.