A Quantitative Model Of Public Opinions' Changes Based On Time Windows

At present,along with the rapid development and popularization of the mobile Internet and smart phones,people are enjoying the great benefits brought by the Internet era,such as: Internet chat,sending and receiving emails,reading real-time hot news,online shopping,and even buying internet financial products,etc.It is due to the active Internet users in the Internet era,their network activities will generate a large amount of network information.The network information brought by these activities includes people's personal opinions or emotions.These views coupled with the developed social networks will affect other people's views on social events,public figures or policies,and may even affect other people's online shopping behavior.Nowadays,the Internet is the main source of people's daily access to information.Therefore,these Internet text data are of great value for both commercial and academic research.At present,the public opinion analysis methods and applications of texts on the Internet are being studied by scholars.In the first and second chapters,we will introduce the current situation and common techniques of text analysis including sentiment analysis.There are three main methods of existing text sentiment analysis: text sentiment analysis based on sentiment word dictionary,a method based on traditional machine learning translated into text classification,and the last one using deep learning.Due to the development of the Internet and the rapid growth of Internet data,it has become increasingly difficult for people to quickly locate,retrieve,and sort out network information.At the same time,the existing research on the characterization of the public opinion value is rather rough,without taking into account the influence of time factors on the public feelings.The dynamic change process also lacks quantitative description,and it is difficult to accurately discover the dynamic process and key elements of public opinion evolution.In this paper,we will use the emotional word dictionary combined with traditional machine learning methods.First,we propose a feasible solution for building a scalable domain emotional word dictionary.Then we will improve the method of quantifying the public opinion value and propose a new quantitative model of public opinion based on the time window.In the model,the static performance of public opinion in different time periods is obtained by means of piecewise linear regression,and then the change of the trend is used to describe the change of the public opinion over time.Ultimately,we can use the last value to make some experimental predictions,such as predicting a trend that only cares about the future of the stock,an explosion at the movie box office or social topics.In practical sense,we can use the segmented linear regressionbased public opinion shift model proposed in this paper to determine the quantified short-term or long-term public opinions values of news,Microblog,or other social media and that is time-seriesrelated public opinion changes.We can also provide corresponding help for the management of Internet financial risks,and this model has considerable practical significance and value in the era of Internet finance.The experiments in this article are mainly based on the A-shares in China.In the experimental dataset,we use the mainstream media in the network,the news in the economic section of the forum,and the high courts from January 1,2015 to May 1,2017.With regard to corporate judgments.At the same time,by comparing with other commonly used algorithms,we construct a complete financial text analysis model to help experts,scholars,and investors to obtain financial emotional information more quickly and efficiently,and provide evidence for investment or research.The experimental results show that the public opinion change model proposed in this paper has higher accuracy and stability than traditional public opinion forecasting.The accuracy of the proposed model is 12.5% higher than that of the traditional model.Compared with the worst performing RF model,the accuracy of the proposed model is 19.1% higher than that model.In the first chapter,we will introduce the current situation of public opinion analysis at home and abroad.In the second chapter,we introduce the related technologies that may be involved in the analysis of public opinion,in the third chapter,we introduce the production of domain related word dictionary.In the fourth chapter we introduce the main part of the algorithm of public opinions' changes,and we introduce the experiment and conclusion in fifth chapter.The sixth chapter is a summary and prospect.