基于超图和样本自表征的谱聚类算法

发布时间：2018-05-06 06:12

本文选题：谱聚类 + 超图　；参考：《计算机应用研究》2017年06期

【摘要】：针对传统谱聚类算法仅考虑数据点对点间的相互关系而未考虑数据间可能隐藏的复杂的相关性的问题,提出一种基于超图和自表征的谱聚类方法。首先,建立数据的超图,得到超图的拉普拉斯矩阵表示;然后利用l_(2,1)-范数对样本进行行稀疏自表征,同时融入超图来描述数据间多层次的相互关系;最后,利用生成的自表征系数进行谱聚类。利用基于超图的样本自表征技术考虑了样本之间复杂的相关性。通过在Hopkins155等数据集上的实验表明,在聚类错误率评判标准下,算法优于现有基于普通图的谱聚类算法SSC、SRC等。
[Abstract]:A spectral clustering method based on hypergraph and self-representation is proposed to solve the problem that the traditional spectral clustering algorithm only considers the data point-to-point relationship without considering the complex correlation that may be hidden between the data. Firstly, the hypergraph of the data is established, and the Laplace matrix representation of the hypergraph is obtained. Then, the row sparse self-representation of the sample is performed by using the L / S _ 2 / 1 / -norm, and the hypergraph is incorporated to describe the multi-level interrelation between the data. The generated self-characterization coefficients are used for spectral clustering. The complex correlation between samples is considered by using hypergraph-based self-representation technique. Experiments on Hopkins155 and other data sets show that the proposed algorithm is superior to the existing spectral clustering algorithm based on common graphs under the criterion of clustering error rate.
【作者单位】：广西师范大学广西多源信息挖掘与安全重点实验室;广西师范学院计算机与信息工程学院;
【基金】：国家自然科学基金资助项目(61450001,61263035,61573270) 国家“973”计划资助项目(2013CB329404) 中国博士后科学基金资助项目(2015M570837) 广西自然科学基金资助项目(2012GXNSFGA060004,2015GXNSFCB139011,2015GXNSFAA139306) 广西研究生教育创新计划资助项目(YCSZ2016045,XYCSZ2017064)
【分类号】：TP311.13

【相似文献】