Research On Network Public Opinion Prediction Based On SVR With EEMD And SFS

With the advent of the information age,the Internet has become a key tool for people to acquire and transmit information.This makes the Internet-based network public opinion have a faster transmission speed and a wider range of influence.Therefore,effective control and prevention of network public opinion can help to maintain a harmonious and stable social environment.Predicting network public opinion trends is one of the key steps to control network public opinion.Accurate prediction of network public opinion can assist the government or relevant departments to monitor it and prepare for controlling public opinion in advance.Therefore,building an accurate network public opinion prediction model has extremely important social application value.In recent years,the application of machine learning methods in network public opinion prediction has received a lot of attention,such as support vector regression.However,the network public opinion trend is complex and may be accompanied by noise.It is difficult to achieve high prediction accuracy by simple machine learning methods.In addition,the machine learning method has more parameters,which need to be adjusted and optimized for specific problems.In view of these shortcomings,this paper proposes three effective improvement strategies based on support vector regression to improve the accuracy of network public opinion prediction based on support vector regression.In view of the noise of network public opinion,this paper uses the ensemble empirical mode decomposition method to decompose the original time series of network public opinion time series into multiple sub-sequences,and adaptively obtains the optimal number of subsequences.Then support vector regression is built for each sub-sequence,and refactored for network public opinion prediction.Different kinds of kernel functions,as well as different parameters,make support vector regression have different fitting and prediction effects.In this paper,multiple kernel functions are integrated,and the parameters are optimized by stochastic fractal search algorithm.A new multi-kernel support vector regression model is proposed for network public opinion prediction.In addition,this paper combines ensemble empirical modal decomposition and multi-kernel support vector regression into a network public opinion prediction model by two cycles.The inner loop uses a stochastic fractal algorithm to optimize the parameters in the support vector regression,and the outer loop adjusts the number of ensemble empirical modal decomposition subsequences.In order to verify the actual prediction performance of the three network public opinion prediction models proposed in this paper,this paper collects Baidu index and Weibo Index data of “One Belt,One Road” and “Rio Olympics”.Then based on this data,each model is built and make prediction.The experimental results show that the three models proposed in this paper are superior to the simple support vector regression method in overall prediction accuracy.Among them,the model combining ensemble empirical modal decomposition and multi-kernel support vector regression has the best prediction effect.