Abstract: The analysis of sentimental orientation can be regarded as a problem of classification on emotional polarity．Under the background of the mass data processing,we proposed a classification approach in terms of sentimental orientation of texts based on LSA(Latent Semantic Analysis) and SVM(Supported Vector Machine),in order to improve the accuracy of the text emotional judgment．On the concept of semantics,we established a space model of "word-document" semantic distance vectors by the latent semantic analysis,and then on account of the privileges of accuracy and generalization of support vector machine,designed a SVM classifier with semantic distance as the input feature vectors.Experimental results validate that our method effectively improves the classification accuracy compared with the traditional SVM method．The classification accuracy rate rises to near 88% on the test set of Web comments with short sentences and explicit sentimental orientation．

Key words: Text processing,Semantic distance,Sentimental orientation classification,Latent semantic analysis