[Ifile-discuss] Re: How to fle "unknown" emails

[clemens fischer]

"Brett Nemeroff" <address@hidden>:

> Is there a way to, for example, if the best fit is less than a <80%
> chance to file it as "Unknown". It would actually need to give it a
> "Unknown" flag. This way when the refile.learn scripts run, it
> learns how to file unknown emails.

when using ifile you will note that the numbers output by "ifile -q"
have various magnitudes.  a metrik analogous to what you want would
have to look at the differences between the first three categories
mentioned.  if these differences are low, you have a message typical
for a number of categories.

i think this would be a major improvement for any bayesian
classifier, not only ifile, but since ifile is the swiss army knife,
the least specialised of the bunch, it should have such an option.
this option would have to look for differences between the topmost
categories falling below a given threshold and simply output a
special category "UNKNOWN" with the vvalue 1.0.

  clemens