Research On Emotional Evolution Of Network Public Opinions Based On Topic-emotion Joint Model

With the arrival of the era of "we media",the media gradually develops towards the direction of personalization.No matter in WeChat,post bar or microblog,every netizen can express their feelings and views on hot events through online news posts.The network has become a window for the expression of public opinions.However,due to the sudden characteristics of online public opinions,it often intensifies the speed of opinion formation and dissemination.A public opinion event plus an emotional opinion can cause a public opinion crisis.To this end,emotion analysis is combined with public opinion analysis.In the early stage of the formation of public opinion,relevant comments of netizens are collected from microblog and web pages in a timely manner to conduct emotion analysis and identify the evolution process and reasons of netizens' emotional attitudes in public opinion events.The government can timely carry out crisis warning and formulate targeted intervention measures to effectively manage public opinion crisis.In view of this,this paper firstly integrates the methods of semantic role annotation technology,tf-idf+k-means clustering,pointwise mutual information(PMI)and emotion dictionary,and improves each algorithm to build a Topic-Sentiment Joint Model,which can simultaneously dig out the thematic emotion Information in the text and conduct emotion classification on the Information;Secondly,according to the language characteristics of comment text in public opinion events,and the life cycle theory is introduced,the model is applied to analyze the emotional evolution characteristics of netizens towards each topic in the different stages of the development of public opinion events;Finally,the "RYB" child abuse incident was used to verify the performance of the topic-sentiment joint model.It was proved that the F value of the model was 14.24% and9.31% better than that of the ASUM model and the JST model respectively,which improved the efficiency and accuracy of the subject extraction and emotion classification.This model not only considers the topic distribution and emotion polarity information in the comments,but also solves the problem of data sparsity that is common in the topic model.The trend of emotion evolution and influencing factors analyzed bythis model can provide an effective basis for the government to formulate corresponding intervention measures.