Hi,
I use Baysian network to model Text categorization.Following is a
part of my network.
Suppose we have R,a,b,c,d. Each is two-valued. Their relations are
R-->a.R-->b, R-->c and R-->d, and I want to know if R is true
when a,b,c and d is observed,
but the value of P(R|a,b,c,d) is always less than P(~R|a,b,c,d). So I
want to describe that when P(R|a,b,c,d)> 0.3, another node is true. Or
somthing else like adding a cost function. But I do not know how to describe
with metaInformation of propability of nodes.
Should I use some Constant node to make P(R|a,b,c,d) larger?
For example, coin a node e, R-->e. let p(R|e=T)=0.8 and P(~R|e=T)=0.5, and
let node e always observed? Is it the common way to do this?
Thank you in advance if somebody could reply to me and thanks
again for people who had help me
before.
I wonder if there are some books about problems and suggestions of
modeling real word applications with baysian network.
Tong.
This archive was generated by hypermail 2b29 : Tue Apr 17 2001 - 10:25:17 PDT