Research Of SVM Kernel Functions In Text Classification

In the field of text classification,SVM is one of the most commonly used classification algorithms.SVM effectively solves the problem of nonlinear classification by introducing kernel technology.As well know,the nonlinear transformation and feature space determined by different kernel functions are also different.Therefore,kernel function is a key factor affecting the performance of SVM classification.At present,there is still no complete theoretical basis for the kernel function selection in SVM.Nowadys,the selection of the SVM-kernel with suitable form has become a key-points in theoretical research.In SVM,the popular used kernels include the linear kernel,the polynomial kernel,the RBF kernel and sigmoid kernel.Linear kernel,the polynomial kernel and sigmoid kernel belongs to the global kernel function,RBF kernel belongs to the local kernel function.This thesis mainly discusses the performance of the above kernels in the text classification,the main work is as follows:Firstly,this thesis discusses the performance of CHI,MI and TF-IDF feature extraction methods in classification.Simulations demonstrate show that comparing with CHI and MI,the TF-IDF feature extraction method has better performance.Secondly,this thesis studies the performance of four kernel function in text classification.Simulations demonstrate show that sigmoid kernel function performs better than linear kernel function and poly kernel function and has good classification performance in text classification.It is not much different from that of the RBF kernel function classification performance.Finally,based on the combined kernel thought put forward by the researchers,a new kind of kernel function based on combination of RBF kernel function and sigmoid kernel function is proposed.The experimental results show that the new combined kernel function improves the text classification performance of SVM.