| Along with its rapid development, Internet is gradually regarded as the media of publishing information and exchanging ideas. What's more, owing to network's openness and sharing, online emerging a large quantity of reviews, either commendatory or derogatory. But these reviews on the Internet often put up their unbalanced distributional characteristics, either with commendatory ones more than derogatory reviews, or to the contrary. So, how to organize, process and analyze their commendatory-derogatory sentiment orientation, has become one of the hot spot of the research field of information science and technology.Aim to commendatory-derogatory sentiment orientation classification of unbalanced review texts, the major study works of this thesis include:(1) In order to deeply analyze the distributional characteristics of the unbalanced review texts, firstly, we made several relative computations on distributional characteristics of the sentiment orientation reviews on books and hotels using statistical methods, then we classify the commendatory-derogatory sentiment orientation of the two kinds of review texts using support vector machine (SVM) as the classifier. The experimental results show that the recall rate of the derogatory reviews (i.e. the minority) is low.(2) In order to improve the recall ratio of the derogatory reviews (i.e. the minority), this thesis research on two kinds of cutting methods, i.e. class border cutting and equal-density cutting. After using cutting methods, we can get a new set of review texts, whose class distributional characteristics tend to balance. Then, we use two feature selection methods, Information Gain and Fisher, to select features. Finally, we discriminate commendatory-derogatory sentiment orientation of the review text using SVM as the classifier. The experimental results show that the cutting methods improved the recall rate of the derogatory reviews (i.e. the minority). |
PDF Full Text Request