I want to use Bayesian Network mechanism for spam filtering. How do you think it should look a proper topology of the network? What about naive Bayes model? (The naive Bayes model is sometimes called a Bayesian classifier)
If you're new to spam filtering, it'd be a good idea to start with something simple like a naive Bayesian classifier. That way you get familiar with the issues involved in handling the data (reading the email, classifying it, storing your lexicon, etc.) without getting too bogged down in the actually classification code. Once you have the basics of your program working, you can go on to more advanced types of filtering.
I found the discussion in the book Ending Spam to be quite helpful.