[1]殷 会,许建华,许 花.基于LS-SVM的多标签分类算法[J].南京师范大学学报(工程技术版),2010,10(02):068-73.
 Yin Hui,Xu Jianhua,Xu Hua.A Multi-Label Classification Algorithm Based on LS-SVM[J].Journal of Nanjing Normal University(Engineering and Technology),2010,10(02):068-73.
点击复制

基于LS-SVM的多标签分类算法
分享到:

南京师范大学学报(工程技术版)[ISSN:1006-6977/CN:61-1281/TN]

卷:
10卷
期数:
2010年02期
页码:
068-73
栏目:
出版日期:
2010-02-01

文章信息/Info

Title:
A Multi-Label Classification Algorithm Based on LS-SVM
作者:
殷 会 许建华 许 花
南京师范大学计算机科学与技术学院, 江苏南京210097
Author(s):
Yin HuiXu JianhuaXu Hua
School of Computer Science and Technology,Nanjing Normal University,Nanjing 210097,China
关键词:
LS-SVM 多标签分类 一对一分解
Keywords:
LS-SVM mu lt-i labe l class ification one versus one decomposition strategy
分类号:
TP181
摘要:
多标签分类是指部分样本同时归属多个类别.基于数据分解的算法因训练速度快、性能良好而得到广泛的应用.本文采用一对一分解策略,将k标签数据集分解为k(k-1)/2个两类单标签和两类双标签的数据子集.对每一训练子集统一用LS-SVM模型建立子分类器,当出现双标签样本时将其函数值设为0,并确定适当的分类阈值.对情感、景象和酵母数据集的实验结果表明,本文算法的某些性能指标优于现有一些常用的多标签分类方法.
Abstract:
A m ult-i label c lassification prob lem lies in that its sam ples m ay be long to mu ltip le c lasses. Data decompos-i tion a lgor ithms are w ide ly used because of its good perfo rmance. One versus one decom pos ition strategy is adopted in th is paper, and th is strategy decom poses amu lt-i labe l problem into several b inary c lass sing le label or bina ry class double label c lassifica tion sub-problem s w hich can be so lv ed independen tly. For each sub-problem, we build a sub-c lassifier us ing LS-SVM mode l and set the function value zero when the sam ple is double labe ,l then de term ine a proper threshold. Exper im enta l results show that our perform ance is supe rio r to seve ra l ex istent m ult-i label class ification algorithm s w ith som e eva luation crite ria on three benchm ark datase ts Yeast, Scene and Em o tion.

参考文献/References:

[ 1] E lisseeffA, W eston J. A kerne lm ethod formu lt-i labe lled c lassification[ C] / / Proceed ings o fAdvances in Neural Info rma tion. New York: BIOw ulf Techno log ies, 2003: 681-687.
[ 2] SchapireR E, Singer Y. Boo stexte r: a boosting based sy stem for text categor ization[ J]. M achine Learn ing, 2000, 39( 2/3): 135-168.
[ 3] ZhangM L, Zhou Z H. A k-nea rest ne ighbor based algor ithm for mu lt-i labe l classification [ C] / / Proceed ings of the IEEE In ternational Conference on G ranu la r Com puting. H e ide lberg: Spr inger Ber lin, 2004: 718-721.
[ 4] Zhu S H, J i X, XuW, e t a .l M ult-i labelled c lassification using m ax imum entropy m ethod [ C ] / / Pro ceedings o f the 28 th Annual Internationa lACM S IGIR Conference on Resea rch and Deve lopm en t. Salvador: ACM, 2004: 274-281.
[ 5] T rohidis K, Tsoum akas G, Kallir is G, e t a.l M ultilabe l c lassification o f music into em otions[ C ] / / Proceedings Inte rnational Conference onM usic Inform a tion Re trieva.l Ph ilade lph ia: ISM IR, 2008: 325-330.
[ 6] T soumakas G, Ka tak is I. M u lt-i labe l classification: an overv iew [ J]. Inte rnational Journa l of DataW arehousing andM in ing, 2007, 3( 3): 1-13.
[ 7] Li T, Zhang C L, Zhu SH. Em p ir ica l stud ies on mu lt-i labe l c lassifica tion[ C] / / Proceed ing s o f IEEE Internationa lConference on Too ls w ith A rtific ia l In telligence. W ash ing ton DC: IEEE Com puter Soc iety, 2006: 86-92.
[ 8] W an S P, Xu J H. A m ult-i label classification a lgo rithm based on tr ip le class support v ector m ach ine [ C] / / Proceed ing s o f IEEE International Confe rence onW ave let Ana ly sis and Patte rn Recognition. Be ijing: IEEE ICWAPR, 2007: 1 447-1 452.
[ 9] Suykens J K. Least squares suppo rt vector m ach ines fo r c lassifica tion and nonlinear m ode ling [ J]. Neura l Ne tw ork W orld, 2000, 10: 29-48.
[ 10] Pav lid is P, W eston J, Ca i J, et a.l C om bin ingm icroarray expression data and phy logene tic profiles to learn functional categor ies using support vector m ach ines[ C ] / / Proceed ing s o f Annua l Interna tiona l Conference on Com putationa lMo lecu la r B io logy. Co lumb ia: Co lum iaUn iversity, 2001: 242-248.
[ 11] Bou tellM R, Luo J, Shen X, e t a.l Learn ingmu lt-i labe l scene class ification[ J]. Pattern Recognition, 2003, 37( 9): 1 757- 1 771.
[ 12] Tsoum akas G, Katakis I. M ult-i labe l c lassification: an ov erv iew [ J]. Inte rna tiona l Journal of DataW arehousing andM in ing, 2007, 3( 3): 1-13.
[ 13] Tsoum akas G, Katakis I, V lahavas I. M in ing M ult-i label Data DataM in ing and Know ledge DiscoveryH andbook[M ]. New York: Springer, 2010.
[ 14] 李佳阳. 基于双标签支持向量机的快速多标签分类算法[ D]. 南京: 南京师范大学计算机科学与技术学院, 2010. L i Jiayang. A fast mu lt-i labe l classification algor ithm based on doub le labe l support v ectorm ach ine[ D]. Nanjing: Schoo l o f Compu ter Sc ience and Techno logy, Nanjing No rm al Univers ity, 2010. ( in Ch inese)

备注/Memo

备注/Memo:
基金项目: 国家自然科学基金( 60875001) . 通讯联系人: 许建华, 教授, 研究方向: 模式识别、机器学习和信号处理. E-m ail:xu jianhua@ n jnu. edu. Cn
更新日期/Last Update: 2013-04-02