Research On Sentiment Analysis Of Chinese Comment Text

With the growing popularity and wide application of Internet, more and more people began to express their views, opinions and comments on the Internet. The text contains the public attitudes towards hot issues and consumer evaluation of goods or services which they purchased in. This information has an important value for governments, business organizations and individual consumers. However, the large scale information on Internet, which are scattered, non-standard and unstructured. To this problem, sentiment analysis emerged, extraction of evaluation collocation and sentiment orientation analysis are important basic tasks in the area of text sentiment analysis.Online product reviews by are that consumers to express their views to these goods or services which they purchased on the Internet by the experience and perspective in the form of text. The information has important value for businesses and consumers. However, because evaluation statement on the Internet is scattered and massive, text sentiment analysis by artificial means is impossible. This paper concentrates on the evaluation statement on the Internet, using nature language processing technology to analysis this evaluation statement on the Internet, so we can get the evaluation collocation and the orientation of the evaluated object.In this paper, the main work and innovations are as follow:(1) To the problem of extracting evaluation collocation, this paper presents an improved method of extracting evaluation collocation based on nuclear sentences, which extracts evaluation collocation by combining nuclear sentences and syntactic dependency. Currently, there are two common methods of extracting evaluation collocation:one method is to use machine learning methods based on the linguistic features; another is a method based on rules or template. Syntactic analysis plays an important role in the both methods. However, because the grammar of most Chinese evaluation text is not normative, the syntax analysis result is unstable and affects the result of extracting evaluation collocation. To this problem, this paper proposed a method based on nuclear sentences, and then significantly improve the normative of the Chinese evaluation text and the accuracy of syntactic analysis. And it also can add the analysis of the parallel relationship among the emotional words and among the opinion targets when dealing with complex sentences, then improve the evaluation collocation extraction rate of recall. Experiment results prove the effect and efficiency of this method.(2) To the problem of evaluated object orientation analysis, this paper uses a method based on semantic weighted sentiment word to analysis the orientation of the evaluated object. Evaluated object orientation analysis belongs to the attribute level orientation analysis, and it basically on the basis of the dictionary. In orientation analysis, especially in the network evaluation statement often appear network vocabulary, however, it is difficult to identify the network vocabulary based on a dictionary. For this problem, this paper uses the common emotional words to replace network vocabulary to effectively solve the problem that network vocabulary is difficult to identify. In addition, to the problem of evaluate statement which contains implicit information, this paper makes the potential evaluation information extraction rules, this method is effective to solve the problem that it is difficult to identify potential evaluation information from evaluation statement. The experimental results show that the proposed approach in this paper can improve the coverage and accuracy, and then verify the validity method in this paper.