摘要
针对传统Apriori算法在标签等级排序中辨识度不高的问题,提出一种基于最小化描述准则(MDLP)Apriori算法的离散Shannon熵值算法。通过在Shannon熵值公式中增加额外参数,并结合自适应MDLP算法,增加Apriori算法对等级排序中分割点的识别能力,从而更加细致地观察标签细微差异;然后,利用改进算法分别在合成数据集和KEBI测试数据集上的仿真实验显示,MDLP-Apriori算法在Kendall系数精度与偏差、分区数量等指标上均要优于对比算法。最后,通过实验给出最小支持度选取标准。
According to the low identification of traditional Apriori algorithm for label ranking,this paper proposed an MDLPApriori algorithm based discrete Shannon entropy for label ranking. By adding an extra parameter in the Shannon entropy formula,and combining with adaptive MDLP algorithm,it increased the ability of Apriori algorithm for recogniting the segmentation point of the lable ranking,which would be more careful observation the label difference. Then,through the experiments on synthetic data set and KEBI test data set with the improved algorithm show that,the MDLP-Apriori algorithm is superior to the contrast algorithm in accuracy and deviation of Kendall coefficient,as well as the number of partitions. Finally,this paper gave the selection criteria of minimum support degree by experiments.
出处
《计算机应用研究》
CSCD
北大核心
2016年第6期1633-1636,共4页
Application Research of Computers
