Interesting review of spam filters (bogofilter included)

[Peter Bishop]

On 24 Aug 2003 at 8:10, Greg Louis wrote:

> On 20030823 (Sat) at 2251:56 -0400, David Relson wrote:
> 
> > I wonder if there's a meaningful way to
> > determine an even better set of "standard" parms ???
> > 
> Probably not.  The values we distribute derive primarily from your
> experience and mine, and if they work well for others, it must be that
> your experience and mine, taken together, are reasonably typical of
> what the average bogofilter user will see.  To improve our defaults we
> would have to assemble a truly representative sample of the planet's
> email -- not an easy task.  It's to be hoped that all the work we've
> put into telling people how to tune bogofilter for themselves will help
> users optimize for their own environments, which is much the better way
> anyhow.
> 

I reckon the main difference between bogofilter and the other Bayesian 
filters are the cut-off settings. Bogofilter seems to have the least false 
positives and hence more false negatives. So it scores well if a high cost 
is associated with false positives - but I guess the others could be 
adjusted in a similar way by changing cut-offs.
.

-- 
Peter Bishop 
pgb at adelard.com
pgb at csr.city.ac.uk