Bayesian Network Structure Learning Method Based On Causal Effect And Its Application

With the continuous development of science and technology and the increasing amount of data,how to obtain useful information from big data has become a research hotspot and difficulty.Bayesian networks are one of the most important tools for representing uncertain knowledge from data,but they still cannot meet some practical requirements in terms of accuracy.In addition,with the rapid growth of the number of literatures,how to intuitively represent the knowledge contained in numerous literatures has become increasingly important.For these problems,this dissertation mainly does the following work:1)This dissertation proposed a bayesian network structure learning method based on causal effect.The priority order of nodes is described quantitatively through the mathematical definition of the priority degree of nodes.The node order was obtained by descending the node priority degree.The bayesian network was initialized by the combination of node order and the maximum number of parent nodes based on mutual information with K2 algorithm.The initial network was modified by mutual information and BDe score to obtain the final learning results.The experiments in this dissertation were carried out on 20 sets of data from the ASIA network and the ALARM network.The results showed that the proposed method was 16% more accurate and 34% better than the MMHC algorithm in standard deviation.2)This dissertation proposed a bayesian network based literature knowledge representation method.Web crawlers were used to crawl through 120,000 key words of literature related to alzheimer's disease.The key words bayesian network were studied by the method proposed in this dissertation.A artificial bayesian network was constructed to measure learning results.Experimental results showed that the literature knowledge representation method based on bayesian network proposed in this dissertation has an accuracy rate of 87.5%.The learning results covered the three hypotheses of the etiology of alzheimer's disease,which provided a new idea and method for bibliometrics.