Sur les spams re�us depuis le 31 octobre sur mon serveur personnel, 1226 spams re�us, 6 passent � travers les filtres bayesiens, 23 sont d�cr�t�s spam � 99% de chances, plus de 1000 ont �t� vus comme spams � plus de 80% de chances. La base de donn�es est faites au fur et � mesure et portent sur les 6241 spams re�us depuis le 20 mai o� j'ai mis en place spamassassin. J'en re�ois 80 par jour dont 2 au plus passent � travers le filtre. Par contre j'ai mis comme poids sur les filtres bay�siens: score BAYES_70 0 0 3.293 3.010 score BAYES_80 0 0 6.200 3.762 score BAYES_90 0 0 5.027 4.002 score BAYES_99 0 0 6.300 4.108
Pour �tre pr�cis, les chiffres sont: cerbere:/var/tmp/Spams# grep -l BAYES_99 * | grep -c "" 23 cerbere:/var/tmp/Spams# grep -l BAYES_90 * | grep -c "" 760 cerbere:/var/tmp/Spams# grep -l BAYES_80 * | grep -c "" 310 cerbere:/var/tmp/Spams# grep -l BAYES_70 * | grep -c "" 111 cerbere:/var/tmp/Spams# grep -l BAYES_60 * | grep -c "" 16 cerbere:/var/tmp/Spams# grep -l BAYES_50 * | grep -c "" 0 cerbere:/var/tmp/Spams# grep -l BAYES_40 * | grep -c "" 0 cerbere:/var/tmp/Spams# grep -l BAYES_30 * | grep -c "" 0 cerbere:/var/tmp/Spams# grep -l BAYES_20 * | grep -c "" 0 cerbere:/var/tmp/Spams# grep -l BAYES * | grep -c "" 1220 cerbere:/var/tmp/Spams# ls | grep -c "" 1226 (sachant que 1220 = 23+760+310+111+16), Qui a dit efficace :-) Fran�ois Boisson On Wed, 19 Nov 2003 12:03:19 +0100 Nicolas Rueff <[EMAIL PROTECTED]> wrote: > Ainsi parla Jean-Luc Coulon (f5ibh) le 323�me jour de l'an 2003: > > > Bonjour, > > > > J'utilise cette fonction. 99% des swen partent ainsi dans le > > r�ceptacle � pourriel. Mais il est vrai que cela consomme quand m�me > > de la bande passante... Ce qui peut �tre pr�occupant pour certains. > > > > Voici ce qu'on peut trouver par exemple dans l'en-t�te apr�s > > traitement par spamassassin : > > > > pts rule name description > > ---- -------- > > 0.1 HTML_MESSAGE BODY: HTML included in message > > 5.4 BAYES_99 BODY: Bayesian spam probability is 99 to 100% > > [score: 0.9998] > > 0.6 MIME_HTML_NO_CHARSET RAW: Message text in HTML without charset > > 0.1 MICROSOFT_EXECUTABLE RAW: Message includes Microsoft executable > > program > > 1.6 NO_DNS_FOR_FROM Domain in From header has no MX or A DNS > > records > > J'ai jamais eu de score BAYES_99. Mon best-of en points total: > > Content analysis details: (16.50 points, 3 required) > X_PRIORITY_HIGH (1.9 points) Sent with 'X-Priority' set to high > ADVERT_CODE2 (0.9 points) Subject: contains advertising tag > ADVERT_CODE (1.1 points) Subject: starts with advertising tag > EARN_MONEY (0.9 points) BODY: Message talks about earning > money > CLICK_BELOW_CAPS (0.5 points) BODY: Asks you to click below (in > capital letters) > SUBJ_REMOVE (0.6 points) BODY: List removal information > EXCUSE_3 (0.2 points) BODY: Claims you can be removed from > the list > FINANCIAL (4.3 points) BODY: Financial Freedom > HTML_WEB_BUGS (0.1 points) BODY: Image tag with an ID code to > identify you > HTML_80_90 (0.5 points) BODY: Message is 80% to 90% > HTML BAYES_60 (1.1 points) BODY: Bayesian classifier says > spam probability is 60 to 70% [score: 0.6434] > HTML_FONT_BIG (0.2 points) BODY: FONT Size +2 and up or 3 and up > HTML_IMAGE_RATIO_02 (0.5 points) BODY: HTML has a low ratio of text to > image area > HTML_IMAGE_ONLY_04 (1.4 points) BODY: HTML has images with 200-400 > bytes of words > MIME_HTML_ONLY (0.1 points) Message only has text/html MIME parts > FORGED_MUA_OUTLOOK (2.2 points) Forged mail pretending to be from MS > Outlook > > -- > .,p**"*=b_ Nicolas Rueff > ?P" .__ `*b Montb�liard - France > |P .d?'`&, 9| http://rueff.tuxfamily.org > M: |} |- H' [EMAIL PROTECTED] > &| `#?_._oH' +33 6 77 64 44 80 > `H. "`"`' GPG 0xDD44DAB4 > `#?. ICQ 97700474 > `^~. > > We are Penguin. Resistance is futile. You will be assimilated. > >
