Dear S.Karthikeyan, > Dear Allori Lorenzo, > > >I have developed a useful concept for making the bayesians to learn.. > > > >On the mail relay server I have a small imap server that has three > account > >configured on it. Spam, No-spam, and Learn. Spam and No-spam accounts > >can be used by any user to collect false positives(NOSPAM) and > no-tagged > >real spam(SPAM). Theese two accounts are the same accounts for all > users > >to reduce the use of ram for the imap service. For personal privacy > after a > >mail arrives into spam or no-spam inbox is taken away from that > account > >and dropped into the Learn account in its separated folder (so the two > >accounts appears empty). The next step is for the system administrator > to > >see if the right mail is in the right folder (to see if users has made > some > >mistakes ;) ). Then he can launch the sa-learn script. > >As regards secirity I'm using Dovecot imap, and theese account cannot > >receive mail from outside you can only drop mails from a mail-client. > >(The good thing is you can decide which users can have this feature). > > > >Do you think it can be interesting to integrate this inside > openprotect? > > > > > Thanks for your suggestions. Your idea is sure interesting and yes, it > will be a nice feature to integrate into OpenProtect. > > Certain things to make sure are: > > 1. The IMAP accounts shouldn't receive mails through SMTP
Allready done. ;) > 2. The IMAP server should be fairly lightweight and secure, because > making users patch the IMAP server frequently will be painful I'm using dovecot imap, it is small and secure (but you never know.. ;) ) > 3. Authorized users(using a username/passwd pair) should copy the > false-positives(ham mail tagged as spam) to Notspam IMAP account > and false-negatives(spam tagged as ham) to Spam IMAP account Sure that's how it works ;) > 4. Mails in Spam and Notspam IMAP accounts should be moved to > another > pair of IMAP accounts, where the admin can view the mails and > make > sure that the users are not poisoning the bayesian database. For > eg, if a user copies a valid ham mail to the Spam IMAP account, > then it could tag the mails from that sender wrongly as Spam in > the future. As I told you in the previous mail the is a learn account (that only the sysadmin is able to use) with two directories called spam and nospam. When a user drag and drop the mail into the real SPAM account (for example), it is automatically removed from there and goes into the Learn/spam directory where the administrator can see if the mail is really spam. the same thing happens with nospam. > 5. Once the admin is certain of these mails, he should move them to > another set of IMAP accounts where they're fed to sa-learn in a > cronjob and then the mails are purged or archived if possible. > 6. Rebuild the bayesian database every 1 day or so to avoid slow > down > due to fragmentation. We can add a line to crontab.. to do that.. i hava mine sa-learn script modified not to rebuild the database every time you learn. (--no-rebuild option). I can add also this script. > > The above usage of three pairs of IMAP accounts seems a little > cumbersome. If someone else has another idea, don't hesitate to suggest > that too. We'll discuss the merits of that method too. I think it is very useful to have the Learn Account because a system administrator can check if the mail is correcly spam or notspam using a mail- client instead of a shell. But if three accounts is too much we can make two: Firss account (for users) with two subdirectories: spam and nospam Second account (for admin) with two subdirectories: spam and nospam The learn direcories are Learn/spam and Learn/nospam. - to move mails to theese accounts i'm using dnotify. Best, Lorenzo > > Regarding the choice of IMAP servers, I'd like some suggestions on your > preferred IMAP server for this bayesian learning. If its is some secure > piece of software and easily installable like djbdns, then there'd be > no > worries regarding keeping the IMAP server current and patching it for > new vulnerabilities. > > >ps=for this things: do I have to write directly to you or in the > mailing- > >list? > > > > > The mailing list is a better medium, as more guys will be aware of what > we're doing and can give their suggestions. But, you can also > personally > mail me, if you feel that is better. :) > > > cheers, > Karthikeyan, S. > > -- > S.Karthikeyan | Ph: +91 (0) 44 52166646 Fax: +91 (0) 44 52079957 > Opencomputing Technologies | http://opencompt.com > Server Side E-Mail Protection. > > ------------------------------------------------------- SF email is sponsored by - The IT Product Guide Read honest & candid reviews on hundreds of IT Products from real users. Discover which products truly live up to the hype. Start reading now. http://productguide.itmanagersjournal.com/ _______________________________________________ Opencomputing-openprotect mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/opencomputing-openprotect
