Prediction Of A-share Index Based On Neural Network And Public Opinion Index In The Post-epidemic Era

Since the outbreak of the epidemic at the end of the previous year,the new crown pneumonia has undoubtedly had a profound impact on the world economy,including China.From snacks and catering,transportation,accommodation and tourism,film and television entertainment have all suffered different degrees of impact.Sudden public health incidents will continue to spread panic among the public.The information transmission interference about the epidemic and the social group panic will have long-term and short-term effects on the financial market.As an emerging market with a sound financial system,China's options and futures system will gradually introduced,investment entities are generally based on individual accounts,and most of them lack professional investment knowledge.Irrational market participants have insufficient recognition of effective information and immature investment concepts,which led to traders' emotional discomfort during the epidemic.Stability in turn affected the stability of the capital market,leading to the occurrence of abnormal situations such as "same rise and fall" that are not easy to appear in the capital market.Therefore,research on the volatility of the stock market based on investor sentiment volatility is of significance.Therefore,this article selects the Shanghai and Shenzhen 300 Index as the research object to represent the index situation of the Chinese stock market,and selects the Baidu index as the public opinion index and the stock-related technical indicators as the in-market indicators to construct the characteristics,and use principal component analysis to analyze The selected features are reduced in dimensionality.First,the labels are tested for financial time series correlation and the fitting of related models,and then an attention mechanism is added on the basis of the traditional RNN and LSTM models,and the corresponding prediction values are generated by learning the weight distribution of the information.At the same time,the corresponding objective function is introduced,and the Adam algorithm is introduced into the optimization method to improve the convergence speed to the optimal solution.When selecting the parameters,the problem of the minimum value of the objective function of the learning rate after multiple iterations of the model is emphatically studied,and the appropriate learning rate,batch size,number of neural network layers,etc.are selected.When the model is fitted,the loss function of the test set is larger than that of the training set,and the model has an over-fitting problem,so a Bernoulli dropout layer is added to the input layer of the model,and the model is compared through the loss function,It is found that the LSTM based on the attention mechanism fits the model best.