Bogofilter accuracy plummets starting around March 10, 2010

David Relson relson at osagesoftware.com
Thu Apr 1 14:59:43 CEST 2010


On Thu, 1 Apr 2010 13:57:30 +0200 (MET DST)
Pavel Kankovsky wrote:

> On Thu, 1 Apr 2010, Jonathan Kamens wrote:
> 
> > I assume I'm not the only one who has noticed that bogofilter's
> > accuracy has plummeted starting around March 10?
> 
> We experienced it here too. But it was a transient problem, it
> returned to normal a few days later when I fed Bogofilter with enough
> material for training (300-400 messages).
> 

These attacks have been around for years.  I can remember messages with
hunks of totally unrelated texts.  Sometimes the text has even been
on subjects of personal interest -- for example archery and the Mary
Rose (a ship of Henry VIII's navy that sank at Portsmouth in
1545, from which more than 3,500 arrows and 137 whole longbows were
recovered).

Over Jan, Feb, Mar my spam per day has averaged 2237, 1768, and 2250.
My unsures per day (totals per month) have been 69, 118, 358.
As percentages, the unsures are 0.10%, 0.24%, 0.51%

So, yes, there's been an increase in unsures, but the percentage is low.

For comparison, the totals per month of false negatives are 25, 39, and
38 -- which makes the short month of Feb the worst.  Part of this is my
son complaining of music/guitar messages which I thought he wanted and
learned that he thinks of them as spam -- so bogofilter is being
re-educated.

Summary: here in Ann Arbor, MI the changes have been small, not big.

Regards,

David



More information about the Bogofilter mailing list