You have a large number of email addresses that are marked as spam senders and needed to be filtered. The idea is to set up a moderately large bit vector initialized to all zero, and map each input email address (as a seed to a number of different random number generators) into a small number of locations in the vector and set these bits to one.

Now, when a new email address comes, the same number of locations will be generated, and if all the locations have a bit value of one, then it is very likely that this new email address is sending spam.