Training SA

[Michael St. Laurent]

>>> Michael St. Laurent wrote:
>>> Okay, I've got the feedback mechanism in place for training the Bayes
>>> engine.  Now for a few proceedural questions.  ;-D
>>>
>>> I have MailScanner set to add the {Spam?} tag to the Subject line and
>>> to make the original message an attachment.  Will either of these
>>> throw
>>>
>>> off the training process?  Is the sa-learn program able to extract
>>> the original message from the attachment and does it know that it
>>> should do so?
>>>
>>>
>> So what do you do when the silly users put all thier spam in the good
>> folder and thier ham in the junk folder an you have automatically
>> already run the sa-leanr stuff on it?
>>
>> This would certainly be a frequent occurance if we implemented
>> this...how do you handle it?
>>
> The whole reason we havent used any manual bayes already is because we
> dont want to create extra work, which i think could cause us.
> You make them clean the dust out of their machine with a straw and
> then you run sa-learn against the offending messages with the --forget
> option. ;-)
>
> If this is a likely occurrence, give the spamchecker user only read
> access (at least for a while) and manually check the folders for
> obvious mistakes. You could even create a 'Forget me' public folder,
> folderdump those messages in a separate directory and run sa-learn
> --forget against them as part of the cron job.
>
> Ultimately the whole point of feedback is that you assume that your
> users know better than your filter. If this is not the case, then
> don't give them access to feedback.

Thank you both.  Now, back to the original question please...

--
Michael St. Laurent
Hartwell Corporation