Fritz, if you are watching this thread:

I think it would be convenient and useful to see the normality info in
the web interface, instead of having to monitor or capture output from
the rebuild script.  Would it possible to deposit that information
somewhere so it could be seen in the web interface?


Greg Wright wrote:
> As far as I understand it, norm should be around 1.
>   

Right, the key to is all is a balance.  1 represents a 1-to-1 balance -
equivalent numbers between the ham and spam corpus.  A positive number
shows how many more times spam words there are than ham words.  A
negative number reflects the opposite - how many more times ham words
there are than spam.

> You have LOTS more spam words to non spam words, and so potentially your
> results in Bayesian might be skewed towards a positive result. 
>   

That's the idea, or the assumption.  The goal is to have both sides of
the corpus full of messages and thus the database full of words.  I
don't know if we have any actual statistics or supporting information if
that means the normaility should actually be 1 - but I have to admit
that (without being a Bayesian mathematician) I assume it should be, or
relatively close.

> See http://assp.sourceforge.net/fom/cache/70.html for what to do about
> it.

That information is only regarding the use of move2num.pl.  It does not
give information on how to provide more balance to your corpus.



-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Assp-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/assp-user

Reply via email to