The VPOP3 spam filter has a couple of mechanisms which can be used to 'train' it to distinguish between spam and not-spam.
The Bayesian filter this looks at words in messages sent and received and their probabilities of being contained in spam and not-spam, and works out an overall probability of the message being spam or not-spam based on the combination of those probabilities.
The Whitelist contains email addresses which are 'trusted'. Messages received from any of these email addresses will be far more likely to be treated as not-spam than messages received from elsewhere.
Note that VPOP3 will never automatically add addresses to the 'Blacklist'. This is because spammers often forge email addresses, so automatically adding entries to the Blacklist runs the risk of adding legitimate senders to that list. Also, it rarely is of any benefit, since spammers rarely re-use their forged addresses, so a big Blacklist just slows down the spam filtering for only minor benefit.