[Spambayes] aging information

G. Armour Van Horn vanhorn at whidbey.com
Mon Feb 17 12:09:49 EST 2003

Tim Stone - Four Stones Expressions wrote:

> 2/17/2003 1:27:10 PM, "T. Alexander Popiel" <popiel at wolfskeep.com> wrote:
> >In message:  <3E509DBA.30411.320099E2 at localhost>
> >             "D. R. Evans" <N7DR at arrisi.com> writes:
> >
> >>Does spambayes have any concept that "the older information is, the
> >>less value it has"?
> >
> >Not intrinsically.  Some few of us who have slightly bizarre
> >installations may have implemented such; for instance, I have
> >a sliding 120-day window that I use for my nightly retrains,
> >so for my purposes anything over 4 months old is forgotten.
> >(The details of my setup are available in the contrib section
> >in BULK.txt.)
> >
> >I have yet to measure the value of this aging process.
> This measurement is important.  I doubt that it actually accomplishes much,
> but we could use some empirical data.  If it really helps, then we should
> include that function somewhere....

I think it could remove a minor impediment to implementing the system, or perhaps
provide a higher comfort level to those considering it, if the system responds to
changes in interests and tastes. The idea that a training decision would affect
the users' incoming mail "forever" might be intimidating to some. If there is a
simple way to derate all token by 1% on every training session, that would
obviously make the installation less of a risk to future communications.

That's a marketing issue, which would be really hard to test in any objective way.
It also seems that a simple aging system could allow a smaller database for a
given level of accuracy, that is something that is actually testable.


Sign up now for Quotes of the Day, a handful of quotations
on a theme delivered every morning.
Enlightenment! Daily, for free!
mailto:twisted at whidbey.com?subject=Subscribe_QOTD

For web hosting and maintenance,
visit Van's home page: http://www.domainvanhorn.com/van/

More information about the Spambayes mailing list