Re: What is the use of Bayesian Network

[EMAIL PROTECTED] (spamcollector) wrote:

> I want to use Bayesian Network as well, because i found that it is one
> of the powerful tool for machine learning, but as I read through some
> introduction about bayesian network. It looks almost identical to
> bayes theorem although one is expressed in graph, so what??
> (confused??)
[...]
> Could anyone tell me what's the differences between filtering spam
> using naive bayes theorem and bayesian network please...

The so-called naive Bayes model is just a special case of general
Bayesian models. Naive Bayes can also be represented as a graph;
it's just not a very complicated one.

The advantage of working with Bayesian networks is that the
graphical representation makes it easier to focus on the 
relations between different propositions (nodes in the graph);
for most people this is more intuitive than an algebraic 
representation.

Also, Bayesian network software usually comes with algorithms
to automatically derive the appropriate form of Bayes' theorem
for the graph in question. The algebraic operations that need
to be carried out vary greatly in complexity with the form of
the network, what is known (evidence), and what is being queried;
but for a large class of Bayesian networks, the operations can
be automatically deduced and applied.

To get started -- I believe there is a link to Bayesian network
software at http://auai.org/. One package that is easy to use is
called JavaBayes. I believe that some people at Microsoft Research
wrote a paper about using Bayesian networks for spam filtering;
one of the authors was Eric Horvitz, I think. Sorry I can't be
more precise. 

HTH
L.O.

[ comp.ai is moderated.  To submit, just post and be patient, or if ]
[ that fails mail your article to <[EMAIL PROTECTED]>, and ]
[ ask your news administrator to fix the problems with your system. ]