--On 9 February 2010 20:09:24 +0100 Jose-Marcio Martins da Cruz
<Jose-Marcio(_dot_)Martins(_at_)ensmp(_dot_)fr> wrote:
But I (and surely others) have another requirement. It could be
interesting to use this mechanism in an active learning context. That
means, user can receive the message flagged as "label required". So the
feedback should include the information : ham/spam.
Yes, the content of the report remains to be discussed. If we were able to
go down the route of labelling the message on the mailstore, I'd like to
see a rich set of messages available for two way information flow,
including the following meanings:
The server filter thinks this message is junk.
The server filter thinks this message is not junk.
The MUA filter thinks this message is junk
The MUA filter thinks this message is not junk
The user has asserted that this message is unwanted
The user has asserted that this message is wanted
The user has asserted that this message is reportable
Perhaps probabilities could be associated with filter results, too.
<AGENT> asserts with <LIKELIHOOD> that this message is <JUNK/NOT
JUNK/PHISH/etc> because <REASON>
Where AGENT might be the userid of the user applying the report, or the
name of the filter software (eg spamassassin, mail.app, thunderbird).
LIKELIHOOD could be based on a spamassassin score or a bayesian analysis.
REASONs could be filter rules (eg spamassassin hits), free text applied by
the user, etc.
Thus, the mechanism could be used to give very specific information to
users, or to operators.
--
Ian Eiloart
IT Services, University of Sussex
01273-873148 x3148
For new support requests, see http://www.sussex.ac.uk/its/help/
_______________________________________________
Asrg mailing list
Asrg(_at_)irtf(_dot_)org
http://www.irtf.org/mailman/listinfo/asrg