基于领域知识的贝叶斯网络结构学习算法  

Domain knowledge based Bayesian networks structure learning algorithm

作　　者：莫富强[1] 王浩[1] 姚宏亮[1] 俞奎[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《计算机工程与应用》2008年第16期34-36,41,共4页Computer Engineering and Applications

基　　金：国家自然科学基金(the National Natural Science Foundation of China under Grant No.60575023);安徽省自然科学基金(the Natural Science Foundation of Anhui Province of China under Grant No.070412064)

摘　　要：针对SEM算法在缺省数据学习中存在精度偏低和收敛速度缓慢的问题,通过将领域知识引入到SEM算法中,提出了KB-SEM算法,该算法首先用D-S证据理论综合领域知识,然后将采集的知识以禁忌表的方式嵌入SEM中来限制和引导算法的搜索路径,缩小算法的搜索空间。实验表明,KB-SEM算法能有效地提高算法的学习精度和时间性能,且能在一定程度上避免主观偏见和数据噪音的干扰。To overcome shortcomings present in the SEM in the presence of missing values data sets such as lower precision and slow convergent speed,KB-SEM is proposed by introducing domain knowledge.A tabu list derived from collecting domain expert knowledge synthesized by using D-S evidence theory is embedded in the SEM to constrain and guide the searching path, and to decrease the searching space.Experimental result shows that KB-SEM can improve the precision and the speed effectively ,and to some extent avoid subjective bias and disturbance of noise in the data sets.

关 键 词：贝叶斯网络 领域知识 缺省数据:KB—SEM 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]