基于SVD的谱聚类算法在维吾尔语水平测试数据分析中的应用  

Application of Spectral Clustering Based on SVD in Analysis for Uighur Proficiency Test Data

作　　者：贾志先[1] 

机构地区：[1]新疆财经大学网络与实验教学中心,新疆乌鲁木齐830012

出　　处：《自动化技术与应用》2016年第1期38-42,49,共6页Techniques of Automation and Applications

基　　金：教育部人文社会科学研究新疆项目(14XJJAZH001)

摘　　要：在谱聚类算法中,采用拉普拉斯矩阵的特征值和特征向量进行聚类,在一些情况下,并不能得到理想的聚类结果。基于奇异值分解的谱聚类算法,比现有的谱聚类算法更有效。给出了一个利用奇异值分解(Singular Value Decomposition,SVD)的谱聚类算法。在基于SVD的谱聚类算法中,奇异向量的选择、聚类个数的选取和高斯核参数σ估计,对于谱聚类的结果有着重要的影响。给出了利用奇异值进行聚类个数的选取和高斯核参数σ估计的公式,以及奇异向量的选择方法。应用基于SVD的谱聚类算法对维吾尔语水平测试数据进行聚类分析,取得了很好的聚类结果 ,可以从中挖掘出更多有价值的信息。In some cases, it can＇t achieve the desired result of the clustering results by using the spectral clustering algorithm with the eigenvalues and eigenvectors of Laplace matrix. The spectral clustering algorithm based on singular value decomposition is more effective than the existing spectral clustering algorithms. A spectral clustering algorithm is presented by using singular value decomposition. In the spectral clustering algorithm based on SVD, the selection of singular vectors, the selection of the number of clusters and the estimation of Gauss kernel parameter 6 have an important impact on the results of spectral clustering. The formula of the selection of clustering number and the estimation of Gauss kernel parameter sigma, as well as the choice of the singular vectors are given. The SVD is applied spectral clustering algorithm to cluster analysis on the data of Uyghur language proficiency test. The desired results of clustering can be achieved and can dig out more valuable information.

关 键 词：谱聚类 奇异值分解 测试数据分析 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]