Research On Analysis And Evolution Prediction Of Network Public Opinion Based On Emotional Tendency

The rapid popularization of Internet technology has a great impact on China’s society, and becomes an important platform for people to understand the world and exchange the views. The Internet and social public opinion gradually fuse at the time of network technology development. And it causes the concept of network public opinion and network public opinion can be true, fast response to public opinion. As the rapid rise of the social platform in recent years, network public opinion largely cited by the basic understanding of the mass, so the emotional tendency and evolution of network public opinion becoming more and more important, which has a great impact on the government administration and decision-making mechanism. So, the network public opinion analysis technology has come into being.Based on the traditional text classification algorithm-support vector machine algorithm(SVM), this paper introduces the specified emotion dictionary and applies it on feature selection for the network of social media. Then this paper structures text tendency classifier and uses the classifier to discriminant blog sentiment polarity(positive or negative). At the same time, we further study of the evolution of network public opinion- heat of public opinion changes – and consider the influence factors of the heat of public opinion. Finally, we verify the feasibility and effectiveness of the method proposed in this paper through experiment. The main content of this paper is described as the following:1. Rearranging and supplementing dictionary based on How Net Chinese dictionary, and trying to build specified emotion dictionary for the network public opinion analysis, which provided the experimental basis for the construction of the network public opinion text classifier.2. Manually labeling of the original experimental data, and doing the work of data preprocessing.3. According to the characteristics of network public opinion in this paper, we applied the sentiment dictionary to text feature selection, and put forward the method of combining frequency method with mutual information- the linear combination of feature extraction method, then selected the feature which meet the conditions and calculated the weight of it, trained the model, and verified the effectiveness of experiments proposed in this paper.4. Using the emotional classifier trained to judge polarity on the whole micro-blog, obtaining the set of negative micro-blog, then using regression model to analyze the evolution of negative network public opinion, and find the influence factor of public opinion heat. Next, analyzing the significance of each factor on the heat of public opinion, establishing multiple linear regression prediction model, and finally analyzing and forecasting the evolution of the negative public opinion and the overall public opinion heat.Experimental results show that the features we selected is more field, representative and combining the word frequency and mutual information method can well characterize the data in the Internet public opinion emotion classification after introducing of sentiment lexicon. The result is better than only use word frequency and mutual information method in feature selection. In the evolution analysis of Internet public opinion, this paper take the driving factors influencing the heat of public opinion as independent variable of multiple linear regression models, analyzing the significant argument of independent variables and whether there is multi-collinearity. Then the model is tested by residual error, and the model is available for prediction. Finally, the paper compares the heat of the negative information with the heat of the overall public opinion information using regression model, and analyzes the evolution law of the time sequence network public opinion.