Re: What's the best way to process missed spams ?

Guillaume Laurent wrote:
> I use spamprobe in 'train' mode. Following what's described in the manpage, I 
> have a script which calls 'spamprobe spam' on all spams which were missed (I 
> don't call 'remove', though, as I recall reading that it wasn't necessary - 
> perhaps I'm wrong). In anycase, this script doesn't seem to help in anyway. 

I don't understand why you say it's not helping.  Based on the 
information you included in your email the message was scored as ham 
before you ran the script and as spam after you ran the script.

Based on the top terms from when the message was originally score as ham 
the reason was that there were terms in the body that were hammy and 
none of the terms in the header were as significant.  Once you corrected 
the classification with "spamprobe spam" the headers became very spammy 
and thus dominated the top terms array.

If you want to see the spaminess of the rest of the terms in the email 
after you retrained on the message you can use "spamprobe tokenize" on 
the message.  It will dump all of the terms in the message along with 
their counts and scores.

For example you could run

   spamprobe tokenize filename|grep "and libraries"

to see how the score for that term changed after your correction.  I 
suspect that it became far less hammy.

The original top terms array had some terms in it that probably should 
have been very spammy.  My guess is that there was a message or two that 
were misclassified as spam earlier and were not corrected.

You are correct that there is no need to run "spamprobe remove" before 
running "spamprobe spam" on a message.  All of the training commands do 
an implicit "remove" on the message before training.  Running remove 
followed by spam should give exactly the same numbers as just running spam.

All the best,
++Brian


-------------------------------------------------------
SF.Net email is sponsored by: Discover Easy Linux Migration Strategies
from IBM. Find simple to follow Roadmaps, straightforward articles,
informative Webcasts and more! Get everything you need to get up to
speed, fast. http://ads.osdn.com/?ad_id=7477&alloc_id=16492&op=click