Sentiment Classification Based On Deep Learning And Its Application In Public Opinion Analysis

At present,with the rapid development of Internet technology and the emergence of social media,people can express their views and opinions on current political hot events through social platforms anytime and anywhere,and the public opinion information grows exponentially.Due to the instantaneous mass and diversity of format of public opinion information,how to capture the public's emotional attitude from the mass of public opinion information,mining the emotional attitude of the Internet users to the current hot spots and refining and analyzing the valuable key information,so as to guide the positive public opinion transmission,is an urgent and extremely important research topic.This paper compares and analyzes the research status of emotion classification of natural language processing at home and abroad and the improvement ideas of several deep learning neural network models,studies the previous analysis ideas and achievements of public opinion of Chinese text network,and on this basis,analyzes the Chinese text emotion analysis algorithm based on deep learning and its public opinion score applied to a specific current hot event Analyze and study.In this paper,first of all,the theory of deep learning is elaborated,and the mathematical formulas of two typical emotion classification algorithms,CNN and LSTM,are analyzed;three deep learning emotion classification models,CNN,LSTM and bilstm,are constructed to realize the text preprocessing,Chinese word segmentation,word vector and other operation links for Chinese text,and through the Chinese emotion ternary classification standard data set After training,this paper introduces the classification evaluation index and makes comparative experiments and results analysis on three emotion classification models,which proves that the model can effectively solve the problem of Chinese text emotion classification and achieve high accuracy.For the application of network public opinion analysis,firstly,crawling and sorting out the comment text data of the empirical case "Changchun Changsheng problem vaccine event",and manually marking the emotional polarity;through the analysis and comparison of the three emotional classification algorithms in the standard data set,real sample data set and mixed data set,the emotional polarity of public opinion information is better predicted,so as to construct This paper constructs a public opinion prediction model of deep learning emotion classification for specific news events;proposes to introduce statistical time series model based on the application of deep learning algorithm,and introduces the process of ARIMA time series prediction model modeling in detail,taking the accuracy rate as the standard to measure the quality of public opinion prediction model of emotional classification for specific news events,and counts the accuracy rate of each time node As the original time series,the value is processed by difference and time series preprocessing,and the model and prediction of the obtained stationary non white noise series are carried out,and the results of four groups of prediction experiments are compared and analyzed,and the analysis conclusions are drawn.The prediction results can better reflect the general trend of emotion classification public opinion prediction model to capture the public emotion,and also basically conform to the typical characteristics of time series prediction.