Is ESET's spam filter supposed to learn?

I used to use PopFile for spam filtering and it was great in that every time you told it to reclassify an email it had marked as spam (or not spam) it learned from the act. After a couple weeks it grew to be EXTREMELY accurate in determining what was spam and what was not. Unfortunately, it was also very slow and tended to time out. That's why I started using Smart Security's spam filtering.

I had thought that when I select an email message that had not been filtered and selected "Reclassify selected message as spam" that ESET would learn from this and become more accurate. But that doesn't seem to be happening. Is it supposed to learn? If not, then there's no point in reclassifying spam that's missed. Just delete it and be done with it.

 

The Antispam module detects unsolicited email in two ways. The first is through a basic set of rules that are designed to automatically filter unwanted email messages. These rules are included by default in the installation of the program, and are updated along with the virus signature database updates. The second way that the Antispam module detects unsolicited email is by means of the Bayesian filter. The Bayesian filter can be trained on a per-user basis; the user manually marks a sufficient number of incoming emails as legitimate messages or as spam.

here

 

I've marked hundreds of messages as spam and it hasn't gotten one iota smarter. With Popfile it took only a few days for it's Bayesian filter to get extremely accurate. I've been using ESET for months and it's no more accurate than it was on day one.

 

Bayesian spam filtering is a very effective form of email filtering used by almost all antispam products. It is able to identify unsolicited email with a high degree of accuracy. The Bayesian filter can be trained on a per-user basis.

The functionality is based on the following principle: In the first phase, the process of learning takes place. The user manually marks a sufficient number of messages as legitimate messages or as spam (normally 200/200). The filter analyzes both categories and learns, for example, that spam usually contains words “rolex” or “viagra”, and legitimate messages are sent by family members or from addresses in the user’s contact list. Provided that a greater number of messages was processed, the Bayesian filter is able to assign to each message a certain “spam index” and thus decide whether it is spam or not.

This is from here. I'd contact tech support in your case