Spammers VS Bayesian Filters

In a previous article I discussed what a Bayesian filter is, and why they are one of the most effective tools against spammers. But it's important to realize that the war against spam is just that - an ongoing struggle. Spammers must change their tactics, and anti-spam software must change it's own tactics to keep up, and hopefully even stay ahead of the spammers. In any such fight it is ultimately adaptability and versatility that win out.

Let us quickly summarise how a Bayesian filter works. Rather than outright banning words, Bayesian filters assign scores to them based on how likely they are to be found in Spam Messages. The more messages that are either accepted or rejected by the user, the more accurate the Bayesian filter becomes in predicting whether any new message is a spam message or not.

However Spammers aren't silly. I have tried to stress this in my articles on the subject. Annoying - yes. lacking in morals or consideration - certainly. But they aren't stupid. And as a new counter to spam comes out, so does a new avenue of attack by the spammer.

Bayesian filters developed from more simple filters that blocked out certain words. Take the word 'Viagra'. Because it is usually picked up by even the simplest spam filters, spammers have taken to writing it as 'v1agra', 'vi4gra' or any of a number of different variations to bypass the spam filters. Now the initial result of this is slipping past a filter that blocks only 'viagra'. The more long term result is that the word 'v1agra' gets an even higher 'spam score' from Bayesian filters than the original. After all, anyone sending you an e-mail using it is almost certainly writing it that way specifically to fool your anti-spam filters. It may slip through once or twice, but ultimately it is more of an alarm bell to see the modified version of the word than the orginal.

This same rule applies to a multitude of methods of getting past filters. In HTML messages 'Viagra' and 'Viagra' will actually look the same, with the HTML simply ignoring the unrecognisable tags and printing everything it sees in between them. But once again the telltale sign of a word deliberately trying to trick a filter becomes a warning bell in itself.

A further annoying spammer trick is sometimes called 'Bayesian Posioning'. Have you ever received an e-mail that contains a paragraph or two of some strange story that makes little or no sense by itself? This sometimes confusing behaviour is actually an example of spammers fighting back against Bayesian filters. Think for a moment what you did with those messagesmark them as spam? Now think about how the Bayesian filter works. Every time you mark a message as spam it analyses the words in it and increases their 'spam score'. But with these Bayesian Posioning letters you are adding to the spam scores of words used in a perfectly normal context. These spam letters aren't meant to sell you anything, they are just meant to soften up your bayesian filters for a future spam attack. The spammers have made themselves doubly annoying. Not only are they sending out their unwanted commercial messages, but now they send out unwanted nonsense letters as well!

The above paints a pretty bleak picture. How can we fight spammers when they constantly change the tactics that they use to make our filters and defenses ineffective? Don't be downheartened. Spammers aim at the lowest common denominator. Most of the people who are overwhelmed by spammers muck have little idea what is happening or why. The mere fact you are reading this article indicates you do, and hopefully now you have a slightly better understanding of how bayesian filters work, what they can and cannot do and how Spammers attempt to defeat their usefulness.

Article Source: http://www.content.onlypunjab.com

With an estimated 70-90 billion spam messages sent every day, the problem is not going away. Don't wait for someone to solve the problem for you, visit The Stop Spam Now Site and review the very best methods of stopping spam.

Please Rate this Article

5 out of 54 out of 53 out of 52 out of 51 out of 5 

Not yet Rated