Fusion Method, Based On The Rating Of The Bayesian Network

It is well known that people always mine useful information from existing knowledge, acquire and make inferences on the knowledge highlighted by many real world applications. This can provided reasonable basis for the real-world problems, such as analysis, prediction, decision and so on. When these knowledge provided by different experts or systems, we should represent knowledge in a unified way and integration of them for analyzing an issue from a strategic and comprehensive perspective. Knowledge always has uncertain and usually contains some causal relationships. Bayesian Network (BN) was proposed to sloving uncertainties of expert system. The DAG represents conditional dependencies. CPT takes as input a particular set of values for the node's parent variables and gives the probability of the variable represented by the node. Recently, BNs have also been successfully applied into a variety of problems, including knowledge discover, decision and prediction, data mining, and so on.Therefore, in this paper we focus on fusing BNs oriented to knowledge fusion. In order to simplify our discussion, we only consider the case that the concerned BNs have the same set of discrete variables. Based on the existing Bayesian network fusion and inference mehod, we propose the BN fusion method without regard to data sets. Preliminary experimental results show that our method is not only feasible but also efficient.The main contributions and novelties of this thesis are summarized as follows:Based on the union and intersection of those BNs, we first construct an initial structure and edges that are escaped from the initial structure. Based on the concepts and properties of BN, we obtain a series of candidate models from adding those edges into initial structure. We then select the final model with the highest score as the final result. The above work solved BN structure fusion problem. Based on the BNs and mature reasoning methods, we can obtain the final results of the parameters. Also, in this paper we proposed some optimization strategy and fusion algorithm, we propose the BN fusion method without regard to data sets, and all the required parameters can be obtained by the mature Bayesian inference method.