The document compares a naive Bayesian classifier and keyword-based anti-spam filtering using personal emails. It describes building a naive Bayesian classifier to classify emails as legitimate or spam based on the presence of words and phrases. The classifier was tested on 1789 personal emails labeled as 211 legitimate or 1578 spam emails. The classifier's performance improved when additional attributes like phrases and non-textual properties were included. It was also tested on a separate corpus of 481 spam and 618 legitimate emails, using cross-validation to evaluate accuracy.