Talk:Bayesian poisoning

From Wikipedia, the free encyclopedia

You have new messages (last change).

[edit] Contradiction

According to this entry:

If spam filters are trained using these emails, there is a much greater chance that they will also mark incoming non-spam emails as spam, due to similiarities between those and the literary sources used to generate the poisoned emails.

According to the external link http://www.zdziarski.com/papers/boudville.txt

No, it won't increase false positives - it will simply cause the filter to look at different characteristics of legitimate mail.

Would someone like to reconcile these conflicting viewpoints? I am inclined to believe the latter source, as it is not anonymous and posted on a world-editable piece of shit website. -76.209.56.104 11:34, 23 November 2006 (UTC)

Retrieved from "http://en.wikipedia.org../../../b/a/y/Talk%7EBayesian_poisoning_e366.html"