Active Learning Method for Multi-label Classification Based on Average Expectation Margin
Duan-Yang Liu, Weijie Qiu · Jisuanji gongcheng · 2011
Aiming at the problems that active learning in multi-label classification is slowly,this paper proposes an improved method for multi-label classification which based on average expectation margin.The method by calculating Support Vector Machine(SVM) expectation margin as the selection criteria.Experimental results show that method proposed in this paper outperforms than other active learning strategy based on decision value and posterior probability strategy in terms of classification accuracy or Hamming Loss or Coverage.It can evaluate the unlabeled sample more appropriate,increase the classification accuracy and classification rate more efficient.