h-MMHC算法及其在主因素分析中的应用

H-MMHC ALGORITHM AND ITS APPLICATION IN PRINCIPAL FACTORS ANALYSIS

由于MMHC算法是针对所有的属性进行的网络结构图的构建,时间相对较长且结构图较为复杂。针对该情况,提出了启发式h-MMHC算法。它是MMHC算法的改进,从一个初始的属性集合出发,通过MMHC局部学习方法,借助启发策略,逐步添加新的属性,最终得到属性之间相关关系的贝叶斯网络结构。该研究以教学效果评估为实例,对于MMHC和h-MMHC算法做了比较。采用李克特量表法设计的调查问卷收集数据,使用两种算法对调查数据进行分析。相对于MMHC算法,由于减少了需要考虑的属性集规模,因此h-MMHC可更有效地应用于主因素分析中。

Since MMHC algorithm is a construction of network structure diagram for all properties,its operation time is relatively long and its chart is somewhat complicated.In view of this,we propose the heuristic h-MMHC algorithm,which is an improvement of MMHC.Starting from an initial attribute set,the h-MMHC algorithm utilises MMHC local learning method and heuristic principle to add new attributes incrementally,and eventually obtains the Bayesian network structure of correlation relationship among attributes.Using teaching effect evaluation as a concrete example,in the paper we compare MMHC and h-MMHC algorithms:using the questionnaire designed by Likert scale method to collect data and employing these two algorithms to analyse the surveyed data.Relative to MMHC algorithm,due to the decrease in the size of attribute set to be considered,h-MMHC can be better applied to principal factors analyses.