Learning to filter spam e-mail: A comparison of a naive bayesian and a memory-based approach