Hapax survival over time

[David Relson]

On Wed, 24 Mar 2004 06:57:01 -0600
Bill McClain wrote:

..[snip]...

> Still, this doesn't answer the question "how valuable are the old
> hapaxes?" They are providing some value, but perhaps the messages that
> use them were already very spammy. 
> 
> It would be interesting to know: given a hapax "X", scored as spam
> because of the overall score of its message, how likely is it to stay
> spammy when it is seen again, or instead to drift toward neutrality,
> or even to cross into ham territory? 
> 
> I spent a small amount of time actually examining the hapaxes that are
> eliminated each day, trying to see if the tokens were of a specific
> type. They seemed to be of all types.

>From Oct 2002 through Dec 2002 _all_ my mail was used in training
(through the combined magic of autoupdate and manual handling of unsures
and errors).  Since training changes a token's ham/spam counts and
timestamp, I can look for old hapaxes and know that they haven't been
seen again.  FWIW, I have 

    70,143 hapaxes from 2002 and 
   260,326 from the first half of 2003
   386,290 from the second half of 2003

Looks like many hapaxes do not reappear.