Will “training” spam filter improve the recognition rate of incoming spams?
Yes, BUT (a big but) it tends to increase the false positive rate, too. Many spams are reflected from well-behaved email domains and have generic contents. This renders natural-language-processing-based and sender-based training ineffective. The system rules are updated regularly and they seem quite balaced for most of the cases. We will stick to it, at least for the moment.
