Re: [Ifile-discuss] How much does reindexing a message affect filtering?

[Clemens Fischer]

Jason Rennie <address@hidden>:

> I wouldn't recommend it in general.  I've heard reports of it (inserting a
> message multiple times) being helpful in specific cases (e.g. spam that
> you are likely to see again).  But, the technique can also hurt by
> artificially inflating word statistics.  If it's being done in moderation
> (no message inserted more than 2-3 times) and evenly (no folder left out),
> it may not have much of an effect, but the results could be quite
> disastrous if you take it to an extreme.

this is my experience as well.  inserting messages more than twice or
three times can actually decrease precision.  the benefit depends on
how ifile is used.  for general classification into many folders, when
the "differences" between certain folders aren't "big", ie. the
folders in question share many words in their statistics, the result
is not easy to calculate and depends highly on the message used.

on the other hand, if the difference is high, then it's not of much
use either, because ifile should learn the concrete type of message
easy, so to speak.

> Best is probably to run your own tests and keep an eye out for weird
> behavior.

this is also educating and fun :)

clemens