Research on advanced filtering algorithm for anti-spam based on a Bayesian classification model

Currently, naive Bayesian algorithm for Email filtering has been accepted widely and been applied in many commercial Email systems. However, facing with the still severe situation of spam Email, people are realizing gradually that depending on such a simple algorithm can't meet the practical needs for anti-spam. On the other hand, Bayesian network, as an important branch in the knowledge discovery area, has been researched and explored for a long period. The question of Email filter can be projected to a Bayesian network model. By doing so and utilizing Bayesian parameter estimation for some key nodes in the model, we can accomplish Email classification and discrimination based on probabilistic conditions. According to the experimental results, new algorithm has much higher rate to converge and more stable than the naive Bayesian one.