Cerb performs a simple statistical analysis of the contents of the first message on new tickets to predict whether a message is spam or not.
The predicted probability of being spam is stored as a spam score on the ticket.
The spam predictions improve and adapt over time through training. When a worker replies to a client message, Cerb learns to be more approving of similar messages in the future.
When workers click the Report spam button on a ticket, Cerb becomes more critical of similar messages in the future.