Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (14): 148-154.DOI: 10.3778/j.issn.1002-8331.1804-0254

Previous Articles     Next Articles

Spectral Clustering Algorithm Based on Local Covariance Matrix

DU Tingting, WEN Guoqiu, WU Lin, TONG Tao, TAN Malong   

  1. Guangxi Key Lab of Multi-source Information Mining & Security, Guangxi Normal University, Guilin, Guangxi 541004, China
  • Online:2019-07-15 Published:2019-07-11

基于局部协方差矩阵的谱聚类算法

杜婷婷,文国秋,吴  林,童  涛,谭马龙   

  1. 广西师范大学 广西多源信息挖掘与安全重点实验室,广西 桂林 541004

Abstract: For the traditional spectral clustering algorithm does not solve the cluster division process, the cross-cluster cross-region sample points have an impact on the clustering effect. In this paper, a spectral clustering algorithm based on local covariance is proposed. The algorithm mainly introduces a new method for calculating the similarity affinity matrix between samples. Firstly, it divides the child by calculating the Euclidean distance between sample points. Then it calculates the covariance matrix of the small subset, sets the threshold to eliminate the intersection, the remaining point constructs a similarity matrix, and then the eigenvalue decomposition of the similarity matrix is done, and finally it uses the classical [k]-means algorithm for the eigenvectors matrix clustering. Experiments on real data sets such as Control show that the algorithm of this paper obtains better results in terms of clustering accuracy, standard mutual information and other indicators.

Key words: spectral clustering, covariance matrix, similarity matrix

摘要: 针对传统谱聚类算法没有解决簇划分过程中,簇间交叉区域样本点对聚类效果有影响这个问题,提出一种基于局部协方差矩阵的谱聚类算法,主要介绍了一种新的计算样本之间相似度亲和矩阵的方法,即通过计算样本点之间的欧氏距离划分出小子集,计算小子集的协方差,通过设定阈值剔除交叉点,由剩下的点构造相似矩阵,对相似矩阵进行特征值分解,用经典的[k]-means算法对由特征向量组成的矩阵聚类。通过在Control等真实数据集上的实验结果表明,该算法在聚类准确率、标准互信息等指标上比较对比算法获得更优秀的效果。

关键词: 谱聚类, 协方差矩阵, 相似矩阵