Nonlinearly Assembling Method and Its Application in Large-scale Text Classification

Support Vector Machine (SVM) is one of widelyused text classification method. Although SVM performs well in practice, SVM encounters two problems: the data distribution is not taken into consideration in the process of classification and its performance is greatly influenced by noises. In view of this, Fuzzy Support Vector Machine based on Manifold Discriminant Analysis (FSVM-MDA) is proposed and Web text classification system is constructed based on Manifold Discriminant Analysis (MDA) and the fuzzy technology. The advantages of the proposed method are (1) it takes both the global and local characteristics into consideration; (2) it has the ability of noise-resistance. Comparative experiments on the authentic datasets show that the proposed method performs better than traditional method SVM.