Research On The Application Of W-SVM In The Early Warning Of Network Public Opinion

With the continuous development of big data and the Internet,Microblog,Tik Tok,Zhihu,Kuaishou,Baidu Forum,We Chat and other network media have become important discussion place for people to express their views on various hot issues and phenomena in their lives.On these network place,People can speak freely their viewpoint.however,too much liberalization has also brought some negative effects to the network environment,such as some netizens making false remarks on the network platform,causing panic,revealing users' personal information and human flesh search.Therefore,timely warning of changes in online public opinion is of great significance to maintaining social fairness and justice and building a clean and upright network environment.The main work of this paper are as follows:(1)Using RSBRA algorithm to establish the index system of network public opinion early warning.Firstly,Based on Boolean reasoning efficient algorithm,the complex continuous data of network public opinion is discretized;then,the importance of discernible matrix attributes is calculated by using heuristic algorithm of rough set under the condition of the original information is unchanged,and redundant indexes are deleted according to the simple rules;finally,an optimized network public opinion early warning index system is established.(2)Aiming at the problem that SVM can only classify but not get the class probability estimation,the W-SVM probability estimation model is established.Firstly,on the basis of weighted support vector machine,m classifiers with different weights are established for multi classification problems.Then,the Bayesian rule method is used to classify the sequences to obtain the segmented probability estimation,and the positive class probability estimation is obtained by synthesizing the segmented probability.Finally,the optimal parameters are obtained by empirical GKL minimization,and the W-SVM probability estimation model is established.(3)The empirical analysis of W-SVM probability estimation model in pre warning of network public sentiment.Firstly,based on the application of W-SVM probability estimation model in network public sentiment pre warning,four risk level confusion matrices of network public opinion early warning are obtained by Python software;then,using confusion matrix as a tool,the model evaluation indexes such as accuracy and AUC value are set to verify the prediction ability of the model;finally,the accuracy is more than 90% and AUC value is greater than 0.85,which proves that the model has good prediction ability and is suitable for network public opinion early warning.