On 6/14/04 11:41 AM, Peter C.S. Adams deftly typed out: >> No, the JMF does not learn. Already junk mail is learning how to beat the >> Bayesian filters that have become quite popular very quickly. It won't be >> long before they loose their early effectiveness. > > Barry, do you have any references to support this? I've never heard anything > about spam "beating" Bayesian filters.
Spammers are now employing techniques dubbed "Bayesian poisoning". They will insert random words or passages of text in their messages to cause their message to score differently than it would otherwise and "poison" your Bayes database. It's debatable whether this would actually decrease the effect of a personal Bayes database. First of all, the use of random words will have very little effect a Bayes database since it is very unlikely that valid mail would have such a random collection of words. The use of passages of text is better thought out, but the spammer would have to choose a passage of text that would be something that would be considered ham based on your selection of mail. Since your collection of spam and ham is based on your individual selection this means that the spammer would have to either find a magic one-size-fits-all paragraph or individualize each and every message. A determined spammer can craft their messages over and over until the message gets through the filters for a broader audience, but that is not worth the effort to attempt on an individual level. -Remo Del Bello -- "It's hard to be religious when certain people are never incinerated by bolts of lightning." - Calvin -- To unsubscribe: <mailto:[EMAIL PROTECTED]> archives: <http://www.mail-archive.com/entourage-talk%40lists.letterrip.com/> old-archive: <http://www.mail-archive.com/entourage-talk%40lists.boingo.com/>
