Why the Random Text?

A large portion of spam these days seems to have large blocks of randomized text included. No, spammers aren't trying to be avant-garde, there's a technical reason behind the computer generated poetry.

Many anti-spam programs utilize Bayesian spam filtering techniques. Bayesian filtering is based on predictive statistics. By analyzing large numbers of spam emails, a database is generated of phrases that commonly appear. How frequently these key phrases appear relative to the rest of the email generates a score indicating how likely the message is to be spam.

Since the indicative phrases are considered relative to the entire message, the randomized junk text can throw off the filter.

SpamButcher's fuzzy logic spam buster isn't as susceptible to this strategy. It does consider the size of the message as a data point. The extra text does skew the filter results, but only to a limited degree. Fuzzy logic isn't as heavily tied to machine learning as Bayesian filtering is, and hence avoids some of its problems.

Bayesian filtering can be entirely emulated within a fuzzy logic system by using the correct rules. SpamButcher's antispam software adopts some Bayesian concepts and benefits from a combination of human and machine training.

Back