A Multi-level Classification Method For Public Opinions In Network Platforms

Network public opinion refers to the network public opinion with different views on social issues that are popular on the Internet.It is a form of social public opinion and a collection of opinions,polarities and emotions expressed by people on objective subjects.The identification of topics,emotions and other public opinion information contained in the network platform can provide network public opinion supervisors with more information supporting management decisions.However,the multi-source heterogeneous network platform does not have a unified classification system of public opinion,and the mined topics and sentiment and other public opinion information cannot be well integrated.How to construct a public opinion classification system suitable for the network platform and accurately identify the abstract topic and emotion information in the network platform are the problems that this paper studies and solves.To solve above problems,we design a theme based on the improved level of a semi-supervised clustering mining method,in view of the heterogeneous network platform for theme mining,to extract which contains the hierarchy information between subject and theme,to build public opinion multistage support topic hierarchy in the classification system,guarantee the timeliness of a subject classification system.A multi-level analysis model of public opinion on the network platform is designed to analyze the topic and emotion level,identify the topic information in the data according to the existing topic classification system,realize the emotion classification model based on the topic characteristics,construct the emotion level in the multi-level classification system of public opinion,and identify the emotion information in the data.Finally,multiple models are integrated to build a multi-level classification system for public opinion,which provides a complete function from data acquisition on the network platform to the presentation of public opinion information such as theme and emotion.In order to verify the effect of the theme mining model in this paper,high-quality theme data from existing platforms are selected to construct a verification set,which is compared with a variety of theme mining methods.Experiments show that the theme mining method in this paper has certain advantages in accuracy and validity of theme level information.In order to verify the effect of the multi-level classification model of public opinions in this paper,experiments were carried out at the two levels of topic and emotion respectively.Verification sets were constructed by manual annotation and collection of common corpus to verify the effectiveness of the topic classification system.Meanwhile,the emotion classification model in this paper was compared with a variety of emotion classification methods.The importance of subject features to the task of emotion analysis is verified.The experiment shows that the emotion classification model based on subject features in this paper is superior to the compared emotion analysis model in classification accuracy.