Research On SVM Text Sentiment Classification Based On Optimization Of IG And RBF

In recent years, network information has growth rapidly with the development of the IT industry and Internet technology, meanwhile, President Xi Jinping put forward the concept of the Internet plus, therefore, a large number of text messages produced by Internet has research value, for example, Internet users will publish some subjective information that can express their thoughts, such as standpoint, suggestion, emotion and so on. Mining and research on these data is of great value. At present, there are two mainly methods of text sentiment classification, one is based on semantic understanding, the other is based on machine learning, and this paper research on the related methods and algorithms based on SVM.Following three aspects would be studied in this paper:First, in-depth study on IG selection method, and proposed relevant solutions to solve the existing problems in the present.This paper has discovered that the feature selection methods ignore the influence of the feature item distribute in the category or among the category on feature selection through literature reading and research, and this result in the feature selection inaccurate. Therefore, based on the traditional feature selection method, this paper proposed add two calculation factors, the characteristic frequency in category and the characteristic frequency among categories to make feature selection more reasonable, applied this improvement to text sentiment classification to improve the efficiency of classification, and verified the classification results by experiment.Second, study on the kernel functions of SVM, fine tune the commonly used Radial Basis Function, and applied it in Combination function.This paper has discovered kernel functions has great influence on the result of SVM text sentiment classification after study on text sentiment classification based on SVM, has found RBF has a very good performance and Combination function performs better than Core function after further study on several common kernel functions. However, RBF has several problems that the data away from test point is too scattered and it's learning ability decay too fast, so linear weighted combination of the improved RBF with polynomial, and with Sigmoid based on the idea of combinatorial kernel to improve performance and accuracy of the classification.Third, this paper optimized the traditional SVM text sentiment classification model based on the above two points, and the result of the experiment shows that the optimization method of SVM text sentiment classification method proposed in this paper has better classification performance and accuracy.