User-guided knowledge discovery using Bayesian networks.

A Bayesian network is a graphical model that encodes probabilistic relationships among variables of interest. When used in conjunction with statistical techniques, the graphical model has shown to be remarkably effective for some data-modeling problems. In this paper, we represent a computational model to apply Bayesian networks to knowledge discovery under uncertainty in a decision support system. Major features of this model include user-computer interaction and iterative information extraction. The user plays a primary role when determining the acceptance or refusal of intermediate information, while the computer in a supporting role crunches the numbers. Two computation streams are provided in the model: (1) Top-down stream: the user enters the expectation value for the goal, and then calculates the expected values for all the nodes in the network. (2) Bottom-up stream: the user input provides evidence into the network, and testifies the effect of the evidence to the goal node. We also designed and developed a software prototype to demonstrate the application of the proposed model. By using the software prototype, the user can easily construct and modify a Bayesian network. Not only does the network establish a connection between the customer requirement and the given source data, but also serves as the tool for our knowledge discovery process. With a tentative Bayesian network, propagations are carried out to testify the relevance represented in the network. After reviewing the results, the user may decide to remove some irrelevant components from the network, or he may want to add new components into the network, which will start a new iteration of the know ledge discovery process. As the repetition goes on, the user will get closer and closer to reach a Bayesian network that suits the problem domain. The information retrieved by applying the derived Bayesian network, together with the network itself, will then be used in further decision support