Abstract:In practical application of text classification for specific domains, most of the text always dopes with each other and is unable to be expressed in linear. The application of kernel function in SVM can solve the problem of nonlinear classification effectively. Different SVM can be constructed by different kernel function, and the recognition performance is also different. So the key problems of SVM are choosing the appropriate kernel function and optimizing its parameters. This paper constructs a new combination kernel function combined with homogeneous polynomial kernel and radial basis kernel function by linear weighted method based on the character of the kernel function. The combination kernel function has good generalization ability and good earning ability at the same time. The simulation experiment result shows that the precision rate, recall rate and comprehensive classification rate of macro average of combination kernel function are obviously better than linear kernel, polynomial kernel and radial basis kernel in choosing the right parameters, and the precision rate and the recall rate are ideal.