Abstract: The efficiency of Bayesian networks structure learning algorithm based on data learning is low due to the big search space. Domain experts determine the inherent causal relationship in network structure according to their own experience knowledge. By collecting different expert opinion, and using the evidence theory to synthesize, it determines the part of causality, removes some meaningless causal relationship, then uses the common learning algorithms to learn, reduces the search space, enhances the efficiency of the algorithm. The experimental results show that the Bayesian network construction method based on the expert knowledge fusion limits the search terms of the learning algorithm by expert knowledge, effectively reduces the search space, uses the evidence theory to synthesize much expert knowledge, prevents a single expert subjective and one-sidedness, can effectively improve the learning efficiency.

Key words: Bayesian network, expert knowledge, causality, evidence theory

摘要： 基于数据学习的贝叶斯网络结构学习算法因搜索空间大而效率低。领域专家可根据自己的经验知识确定网络结构中固有的因果关系。通过收集不同专家的意见，并利用证据理论进行综合，确定其中的部分因果关系，去除其中无意义的因果关系，然后利用常用的学习算法进行学习，减小搜索空间，提高算法效率。实验结果表明基于专家知识融合的贝叶斯网络构造方法利用专家知识来限制学习算法的搜索条件，有效地缩小了搜索空间，利用证据理论综合多个专家知识，防止了单个专家的主观片面性，能够有效地提高学习效率。

关键词: 贝叶斯网络, 专家知识, 因果关系, 证据理论