一种迭代加权的元样本稀疏表示中文文本分类算法

An Interactive Weighted Metasample Based Sparse Representation Chinese Text Classification Algorithm

为进一步提高文本分类的准确率和鲁棒性,在元样本稀疏表示分类算法的基础上,提出一种迭代加权的元样本稀疏表示文本分类算法,该算法在每一步迭代中依据一定的规则有监督地对权系数进行调整,使目标函数值被限定在较小的范围内,逐步逼近最优拉格朗日乘子,以得到更加稀疏的样本表示系数。实验结果表明,与经典的文本分类算法KNN、SVM及非加权的MSRC算法相比,提出的文本分类算法具有较高的准确率和较好的鲁棒性。

In order to improve the accuracy and robustness of text classification, this paper proposes an interactive weighted metasample based sparse representation text classification algorithm on the basis of metasample based sparse representation coding. In each iteration step, the method can adjust weight coefficient through supervision according to certain rules, which make the value of objective function be restricted to a small range, and make the weight coefficient gradually approach to the optimal lagrange multi- plier to obtain a more sparse sample representation coefficient. The experimental result shows that the proposed classification algorithm has better robustness and higher accuracy than the classical text classification algorithm KNN, SVW and non-weighted MSRC.