A SVM Text Classification Approch Based on Binary Tree

Support vector machine(SVM ) is based on minimal structure analysis principle, it can it can solve the dimension disaster, regionally minimal problems, etc. But the common SVM can only solve binary classification. Some research develope algorithm that can solve multi-class classification through constructing binary tree with several binary SVM, the research yields some fruits. Linguistics research result show that of all the extracted feature word, noun and verb make up a great proportions, about 65.5%. Based the above knowledge, we improve the SVM multi-class classification by introducing an algorithm of constructing binary tree, which use the Chinese part-of-speech information to reduce the dimension; we also optimize the binary tree node sequence by calculating the distances of the classes. Experimental results shows that the proposed SVM-multi-class classification have high precision and recall rate.