基于非负矩阵分解的中文倾向性句子识别

Identification of Chinese opinion sentence based on NMF

提出一种基于非负矩阵分解(NMF、SNMF和WNMF)的中文倾向性句子识别算法.该算法首先构建倾向性特征矩阵,然后通过NMF、SNMF和WNMF算法分别来降维、提取潜在语义,最后采用支持向量机分类器识别中文倾向性句子.实验结果表明,与PCA和SVD相比,NMF、SNMF和WNMF算法能有效地降低维度、提取潜在语义,并提高倾向性句子识别的精度.

This paper proposes a new method for the identification of sentence opinion based on non - negative matrix faetorization （NMF, SNMF and WNMF）. The method constructs feature matrix, then applies NMF, SNMF and WNMF to reduce dimensionality of matrix and extract the potential latent semantic information, and finally uses SVM to identify opinion sentences. The experiments shows that, compared with PCA and SVD, NMF, SNMF and WNMF can not only do better in reducing dimensionality and extracting potential latent semantic information, but also can get higher accuracy.