Hi Pierre, > I don't say that (sorry, I'm belgian and my english is not very good).
It's OK, I'm American...my English is not very good either <g>. > I want to know why you CAN place Bayesian algo on SERVER-SIDE (you must > analyse mail incoming on the server but if the mail is crypted or > person who > send mail does not want that you open his mail for analyse,....). The advantage of putting this code on the server side is so that ALL clients can benefit from the SPAM blocking, and you "can" avoid the client even having to see the junk email...which is the whole point in most cases. Also, there are a lot of different email clients out there, and building a custom SPAM blocker for each one isn't feasible. Centralizing this process makes more sense. Plus, we still have the ability to give each user their own custom corpus, rather than having everyone use the same one. However, if you use a different email server (non-James), but still want the features this code provides, it would not be hard to create a Pop wrapper to filter messages as they're downloaded from various servers. This wasn't my goal, but the code lends itself to a variety of implementations. If the message is encrypted, then the SPAM blocking won't work, but there again, what spammer sends encrypted email? As for the other case, where a user doesn't want his/her mail analized... 1. If the message itself isn't being stored in a special Ham repository, only the words/tokens will be saved...with no way to know what message they came from. 2. A mailet entry could be added to exclude that user from from the analysis completely. > Also, what do you do with mail that they are declared spam??? > Spoolmanager place them in the repertory Spam???? > If yes, and after if it's a ham????? It's up to the James administrator/user. You can just add a "MessageIsSPAM: Yes" header to the message, and let the client handle it (filter it into a specific folder or whatever). Additionally, you could add another mailet to persist it to a db table or whatever. The results of the analysis just modify the message and indicate it's spamminess, what you do with it after that is limited only by what a James mailet/matcher can do...which is to say...anything you like <g>. As to a message that was flagged as Spam but the user decided is Ham, I included classes and sample configurations that show how messages could be sent to James and targetted for either the Ham or Spam repositories. Hopefully, this has answered your questions. But feel free to ask more. -Chris > > Pierre > > > -----Message d'origine----- > De : Chris Means [mailto:[EMAIL PROTECTED]] > Envoy� : mercredi 29 janvier 2003 04:38 > � : James Users List > Objet : RE: Spam filtering mailets wanted... > > > Scherer, > > > If I have understood correctly the Bayesian method, you do a global > > probability on the 15 most high probability present words in the mail > > (header & body). > > The client can do that but(he can read the mail because he's > the taregt of > > the mail), on the server side, it's more complicated no???? > (secret of the > > mail,...). > > I don't pretend to completely understand the Bayesian Analysis > theory, but I > believe I've at least come close to implementing it correctly <g>. > > I don't say that (sorry, I'm belgian and my english is not very good). > I want to know why you CAN place Bayesian algo on SERVER-SIDE (you must > analyse mail incoming on the server but if the mail is crypted or > person who > send mail does not want that you open his mail for analyse,....). > > > Blacklist : Why everybody says now that it's not a good solution??? > > I use SPAMCop along with the regular blacklists...to check > messages against > before I bother with the BA. > > > > Chris, you use in your class a package : > > org.apache.james.util.JDBCBayesianSPAMAnalyzer > > Where is this package please???(is it in this one that the > > Bayesian algo is > > implemented???). > > All the code was in the ZIP I emailed to the list. > > Ah?? sorry ;-) > > Pierre > > -Chris > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > _____________________________________________________________________ > Envie de discuter en "live" avec vos amis ? T�l�charger MSN Messenger > http://www.ifrance.com/_reloc/m la 1�re messagerie instantan�e de France > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > _____________________________________________________________________ > Envie de discuter en "live" avec vos amis ? T�l�charger MSN Messenger > http://www.ifrance.com/_reloc/m la 1�re messagerie instantan�e de France > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
