带Spearman相关性的多标签GRF算法  

Multi-Label GRF Algorithm with Spearman Correlation

作　　者：冯元佶[1] 李枚毅[1] 王伟[1] 

机构地区：[1]湘潭大学信息工程学院,湘潭411105

出　　处：《模式识别与人工智能》2010年第6期862-866,共5页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金资助项目(No.60773047)

摘　　要：通过采用Spearman相关系数矩阵取代临时分类标记来构造标签相关性模块,提出一种改进的带Spearman相关性的多标签高斯随机域(MLQ-GRF)算法,以减少临时分类标记的不确定性.实验对比所得结果表明,文中构造的改进的MLQ-GRF算法对于扰动和带误差的临时分类标记有更好的稳定性,能提高分类的精确度.An improved multi-label Gaussian random field algorithm is proposed to reduce the uncertainty of temporary labels.The spearman correlation matrix is used to build a label-relevant module instead of temporary labels.The results of comparative experiments show that the proposed algorithm is stable for temporary labels with tolerance and disturbance and it increases the accuracy of classification.

关 键 词：半监督学习 Spearman相关系数 多标签分类 

分 类 号：TP391.44[自动化与计算机技术—计算机应用技术]