中文摘要：

为了提高贝叶斯分类器的分类性能，Keogh提出了以分类效率为基础的扩展贝叶斯网络分类算法SuperParent—TAN，这是一种依赖一个属性（onedependenceestimator）的贝叶斯网络。这种算法不足之处在于查找超父节点（Super—Parent）和创建分类器工作的反复进行，时间花费较大。为了提高这种算法的分类效率，同时保证分类率，设计了基于信息增益和基于互信息的两种排序算法。通过在Weka平台上对UCI中32个数据集合的实验表明，基于信息排序的优化算法可以在保持分类正确率同时降低分类花费。

英文摘要：

In order to improve the accuracy of Bayesian classifier, classified-based SuperParent-TAN is provided by Keogh to find the set of augmenting arcs, this is a one dependence estimator. But the cost is expensive for the repeatedly finding the SuperParent nodes and creating classifiers. In order to further improve the learning efficacy and efficiency, and manage the accuracy at the same time, two Sorted SuperParent Algorithms are provided in the text. One is ordered by information gains and another is by conditional mutual information. Experimental results on 32 datasets of UCI, indicate the effectiveness of the optimization.