Researched On Sentiment Orientation Of Chinese Text Classification Based On Statistical Method

With the development of the Internet, texts information from the Web is concerned by more and more enterprises and individual. The texts can provide an important basis for the government to understanding people's intention, making policy and improving service. Through the related reports and reviews of some events, we not only learn about the event itself, but also know the people's views, opinions and attitudes etc. Many enterprises have created their own website of products reviews column for making the market research and analysis, knowing the opinions and suggestions of the customers. Through online tracking, the product's quality and service can be improved by products reviews. Product reviews on the Web can guide the behavior of the consumers. However, there are too many views appearing on the web everyday, it is impossible only depending on manual to tracking and analyzing. The people begin to study the subjective sentiment orientation of views text.In this paper, the text sentiment orientation classification has researched for views on the web.(1) Applling the key techniques of text topics classification to text sentiment classificationIn this paper, using three kinds feature selection methods (information gain, mutual information and chi-square statistic), and two kinds probability computer (Boolean and words frequents), and support vector machine to construct the classifier, text sentiment classification are studied. The results indicate that some key techniques for the traditional topic categorization can be used to text sentiment classification. However, the effect is inferior to text topics classification. The reason is that text sentiment classification is more complex on the feature selection and only using category distinguishing ability is not enough.(2) Proposing a words sentiment orientation measuring method based on synonymy sentiment intensity. Words and its synonymy have the same or close sentiment orientation. The association strength of synonymy and paradigm word can intensify the sentiment orientation of object word in some degree. For depicting this language phenomenon, we proposed a words sentiment intensity measuring method which is based on the synonymy sentiment orientation strength. The experiment results indicate that the proposed method is superior to the method based on words sentiment orientation strength.(3) Proposing a method of restricting paradigm word based on category word frequency difference.The classification result of words sentiment orientation depended on the selecting of paradigm words in greater degree. In this paper, we proposed a paradigm words selecting method of restricting corpus. The experimental results showed that this method is superior to the commonly paradigm words selecting method.(4) Proposing a method for identifying the sentiment orientation of combination items based on maximum entropy model.Combination item is one of the important features for the text sentiment orientation classification. In this paper, a method has proposed for identifying the sentiment orientation of combination items based on maximum entropy model.(5) Studying the effects of text sentiment orientation classification using hybrid candidate feature.Using different hybrid candidate features, we studied and compared the results of sentiment orientation classification for the same corpus. The results showed that the more components of candidate feature are., the better of the text classification result is.