zamri wrote:
How di u guys train SA? Logically, I can export mailboxes to mbox format and train them.
Well, I don't do this on a strictly regular basis, but I decided to simplify this process a bit some time ago and forever =) (for me at least).
I use Thunderbird, and I try to examine each message before feeding sa-learn with it (probably because my mail traffic with 50 users is not extremely big). Well, here I mean, that I do care about what messages get to sa-learn, and that's why I decided not to automate all this any more (make per user filtering, let users learn Bayes-filter (and, may be, spoil the base a bit)). So I made local folders Spam and Ham in Thunderbird, where I place the messages to be learnt. These are mailboxes, and I upload them to some particular folders on the server by FTP; cron runs sa-learn 3 times an hour, and writes the result to the log file. These folders on the server (spam and ham) may contain many mailbox-files for sa-learn to process. One folder is for spam and another one - for ham.
I can share the script of mine (as a ready solution), if anyone is interested.
I think, it is not a problem to make a solution that will get messages right from the database (from the IMAP-folder, for example) and give them to sa-learn.
But I don't have such ready solution. -- Best regards, Danil V. Gerun.
