Allen Writes:

Personally I would think twice about recommending SpamBayes. Its false negative rate has been REALLY bad for me. It’s just too easily fooled by chucking in a bunch of random words. At the moment I’m just too lazy to find something better though.

SpamBayes has processed 15355 messages - 14754 (96%) good, 361 (2%) spam and 240 (1%) unsure.

3743 messages were manually classified as good (0 were false positives).

908 messages were manually classified as spam (316 were false negatives).
Recently, one Gnomie emailed me his concern over the user of SpamBayes. It seems like they ere getting a lot of false positives. As a POPFile user, I have not had a ton of sit down experience with this particular spam filtering tool.

Still, I also understand that Bayesian filtering is NOT designed to work out of the box. This filtering requires a fair amount of training, especially considering the fact that spammer are working at throwing a rock into the gears with random characters.

But once you get a handle on what Bayesian filtering is and what effort is involved in helping it learn what YOU consider to be spam, it can be very powerful.

What do you think? Has the concept of Bayesian filtering become entirely too dated? Or perhaps, most of the issues stem from the needs and challenges of specific applications? Email me, let’s talk about it.