Hello Alexander Leschinsky & everyone else, on 07-Dez-2004 at 14:18 you (Alexander Leschinsky) wrote:
> - store only one variation of token instead all possibilities Good for the filter, bad for the user who has to tend to that translation table... > token "Viagra" not identical to "V1ägrä" (for bayes-filters), but it > still readable and understandable for human reading. On creating pairs > you give BIT possibility test not only found token, but also all possible > variation of writing and increase (in ideal) detection quality I've seen the l337 scene type and that oddness is adopted by spammers, yes. You'll never catch all possibilites. On the other hand, if spammers write "V1A6RA" and other variations, the possibility that a message containing a correctly spelled "viagra" will be genuine does increase, so the filter should not be troubled by this (according to the PopFile docs). BayesIt is the first filter to have such a feature, and remembering that all the other Bayes filters (without such a translation table that the user has to tend to - which is definitely too much for the average end-user) worked at over 99% accuracy makes me come to the conclusion that I best leave this thing alone, even though my mail is not all english. :-) -- Best regards, Alexander (http://www.neurowerx.de - ICQ 238153981) using TB! v3.0.2.10 on Windows XP Pro Service Pack 2 The spirit of one individual can supersede and dismiss the entire clockwords of history. (Tom Robbins) ________________________________________________________ Current beta is 3.0.2.10 | 'Using TBBETA' information: http://www.silverstones.com/thebat/TBUDLInfo.html IMPORTANT: To register as a Beta tester, use this link first - http://www.ritlabs.com/en/partners/testers/
