Re: [OFFTOPIC] spam scoring (was: *****SPAM***** Extracting a KeySym from an action routine)

Around 14 o'clock on Nov 26, David Dawes wrote:

> All I can really say so far without having analysed the data is that
> the number of false positives has been relatively small compared to the
> number of valid positives.  I need to assess now many valid positives
> were attributable to the RBL matching before coming to any conclusions
> about that.

However, the false negative rate for xpert has been surprisingly high, 
compared to the results I've seen with spamassassin here at home.  High
enough to prevent me from automatically forwarding messages not tagged as 
spam sent by non-subscribers.

There have been several spam detector papers submitted to this years 
Usenix conference; I'm pretty confident that we'll have better tools 
available in the next couple of months.

We might disable the RBL based rules in flagging spam; they do seem to
have a rather high false-positive rate.

Keith Packard        XFree86 Core Team        HP Cambridge Research Lab