Hybrid of neural network and decision knowledge approach to generating influence diagrams

摘要

Influence diagrams (IDs) have been widely used for a great deal of decision analysis problems. However, specialty is required for generating a well-formed ID which is needed for more complex decision problems. In addition, both decision-analytic knowledge and domain-specific knowledge are required for generating well-formed IDs. To resolve a problem like this, we suggest herein a hybrid approach in which neural network is used for yielding an initial ID representing decision-maker's explicit and implicit cognition about a given problem, and two kinds of knowledge such as decision-analytic knowledge and domain-specific knowledge are applied to refine the initial ID into a well-formed one. An illustrative problem was solved successfully by the proposed approach. Experiment showed that our approach is useful for modeling complicated decision problems with less time and effort in situations when problem is complicated and decision-maker does not have enough resources.