NMF初始化研究及其在文本分类中的应用

Study of Non-negative Matrix Factorization Initialization and Its Application to Text Classification

对非负矩阵分解的初始化进行研究,提出针对文本分类的主成分分析(PCA)、有监督PCA(SPCA)和模糊C平均3种初始化方法并进行了实验。多类文本分类的实验结果表明,这些方法有效地解决了初值对结果的影响问题,不同程度地提高了文本分类结果,其中SPCA优于其他2种方法。

The initialization of Non-negative Matrix Factorization（NMF） has studied in this paper. There are three methods of initialization PCA, supervised PCA（SPCA） and Fuzzy C-Mean（FCM） are reported for text classification. Experimentsal results of multi-class text classification indicate that the three methods effectively solve the problem of results effected by initialized values, and improve the text classification results. The SPCA of the three methods is best.