中文摘要：

贝叶斯网络是一种有效的不确定性知识表达和推理工具,在数据挖掘等领域得到了较好的应用,而结构学习是其重要研究内容之一。经过二十多年的发展,已经出现了一些比较成熟的贝叶斯网络结构学习算法,对迄今为止的贝叶斯网络结构学习方法进行了综述。现阶段获得的用于结构学习的观测数据都比较复杂,这些数据分为完备数据和不完备数据两种类型。针对完备数据,分别从基于依赖统计分析的方法、基于评分搜索的方法和混合搜索方法三个方面对已有的算法进行分析。对于不完备数据,给出了数据不完备情况下网络结构的学习框架。在此基础上归纳总结了贝叶斯网络结构学习各个方向的研究进展,给出了贝叶斯网络结构学习未来可能的研究方向。

英文摘要：

Bayesian networks is an effective tool for uncertainty knowledge expression and reasoning, has been applied in the fields of data mining etc. Structure learning is one of the most important research content. After twenty years of development, there were already some mature algorithms for Bayesian networks structure learning. This paper reviewed the Bayesian networks structure learning algorithms up to the present. The observed data for structure learning were relatively complex, which divided into complete data and incomplete data of two types. For the complete data, it analyzed the existing algorithms which divided into three types, dependency-based statistical analysis method, the method based on the score search and hybrid search. For the incomplete data, it given the learuing processing of incomplete data. On this basis, summarized current research work a- bout Bayesian networks structure learning, and pointed out future research directions.