The structures from the data learning and the experts' experience, which one is better

[wxk8000]

Hello
I am a newbie of bayesian network and GeNIe.I try to make the bayesian networks sttucture learning from the sample data, and obtained the structure. But I found that the structure does not match with the the experts' experience, even some causal relation is just opposite to the experts' experience.
1. The structures from the data learing and the experts' experience, which one is better?
2. How to evaluate the two structure?
3. What to do when encounter this problem?

[marek [BayesFusion]]

It is common to see networks that do not conform to expert's intuition. I typically trust data more than expert opinion but which network is better is an empirical question. The best way to compare two models is to run validation (in Data menu) and see which model gives better accuracy. You can also influence learning by entering Background knowledge. The dialog that pops up when you press the Background knowledge button allows you to force arcs, forbid arc, and to specify temporal precedence between nodes. These constraints will have a large influence on your network.
I hope this helps,

Marek