Research On Text Sentiment Classification Based On The Method Of Combing Optimized Semantic Understanding And SVM

With the rapid development of the Internet technology, information resources aregrowing rapidly. In this case, in order to establish a comprehensive, effective and fastInternet public opinion monitoring and early warning mechanisms, it is more realisticand convenient to use computer technology to automatically collect and analyze thesekinds of opinions instead of manual work. Therefore, main researches of this paperare as follows:First, this paper presents an optimization of semantic understanding. It calculatesthe weight of words in emotion dictionary by using the adjusted HowNet semanticsimilarity calculating method, and then extracts emotion sentences according to thematch pattern of emotion words, and at last, judges the text emotional tendency basedon the integration of multiple dictionaries.Second, this paper combines standard Information Gain、Term Frequency andemotion words degree to select subset features more properly, which improves thespeed of classification and the precision of result.Third, semantic understanding and optimization of SVM are combined to build anew text emotion classifier, it gives full play to advantages of the two methods. Firstof all, labels the samples having high class membership degree using the optimizedsemantic understanding method, and then uses these samples to do machine learning.It does not only improve the model’s portability, but also eliminate the randomness ofselecting training samples.Forth, this paper presents the emotion text classification model on the basis ofthe proposed new method, and the experimental results show that the emotion textclassification method this paper proposed, combing the method of optimized semanticunderstanding and SVM, is feasible.