Research On Key Technologies Of Political Election Prediction Based On Network Information

The rotation of political parties and leaders in major countries or regions in the world may exert important influence on local regions and even the whole world in military,diplomatic,trade,science and technology and other aspects.Accurate prediction of election results in designated countries or regions is an important support for formulating action plans for targeted response strategies.However,election prediction has many influencing factors,great uncertainty and fast information iteration which due to predict elections correctly become harder.Aiming at the problem that the prediction of election from a single source data cannot fully reflect the political tendency of voters,this paper builds a data correction model based on time series,a reverse normalization model,a multivariate data fusion prediction model,a quantified model of netizens’ emotion classification and a prediction model of netizens’ emotion evolution.Analyze the election situation from many angles and aspects.This paper mainly does the following work:1)A prediction model based on polling data correction is proposed.Firstly,the average of the results of several polling institutions is taken as the standard value of the poll results,and the time series analysis method is used to correct the bias series of each polling institution,and a data correction model based on time series is constructed.At the same time,the uncommitted people in the poll data are inferred and analyzed,and the uncommitted data in the poll data are added to the corresponding candidates according to the reverse ratio score of the candidates,and the reverse normalization model is constructed.Then,on the basis of the poll correction prediction model,a multivariate data fusion prediction model is constructed based on the combination of demographic data and historical election result data by using Bayesian framework.2)An election prediction model based on the quantification of netizens’ emotion classification is proposed.In view of the lack of corpus and the inclusion of comment objects in the comment text in the field of political election,this paper constructs the candidate name database,classifies the comment content by sentiment dictionary,SVM and CNN-LSTM,and quantifies the sentiment classification result as the netizens’ support rate to the candidate by using the moving average cumulative probability.The quantification model of netizens’ emotion classification is established..3)A prediction model based on the evolvement of netizens’ emotions is proposed.In view of the timeliness of social media data comments and the complexity of netizens’ emotional changes,this paper analyzes netizens’ attitudes towards candidates in different time periods and their evolution process according to K-means clustering method,LDA model and word cloud construction,and constructs a prediction model of netizens’ emotional evolution.The verification of the real data of the 2020 election results show that,the five prediction models of political election conditions constructed in this paper can all predict the election results well under certain circumstances.Reverse normalization model has the smallest relative error and the best effect.The prediction model of netizens’ emotion evolution can better analyze the influence of netizens’ emotion evolution on the candidate’s support rate and show the change trend of candidates’ support rate in different time periods.