Research On LSTM Stock Index Price Prediction Method Combining News Features

With the development of the financial market,many financial products represented by stocks have been chosen by more and more people.As an attempt to determine the future value of trading company stocks or other financial instruments,stock market forecasting has attracted more and more people's attention.Due to the non-linearity,volatility and complexity of the stock market,it is very difficult to accurately predict the stock index.In addition to various stock data,there are also many news and information that have a subtle influence on investors' decision-making,thereby affecting the trend of the market.Therefore,the realization of accurate prediction of stock index prices can not only enable investors and enterprises to better grasp the trend of stock prices,but also prevent investors from making irrational decisions,realizing timely risk aversion and obtaining income,and has a significant impact on stock investment decisions.Significantly.Based on the analysis of domestic and foreign stock index price prediction methods,this article optimizes the existing stock index price prediction model for the purpose of accurately predicting stock index prices,combines news text data with stock index historical data,and adopts theoretical analysis,model construction and Data processing has realized accurate prediction of the price of the Chi Next stock index.It mainly constructs an LSTM stock index price prediction model that integrates news text features.Through many experiments,the most suitable LSTM model parameters for the input data of the GEM are obtained.The closing price,opening price,highest price,lowest price,transaction amount,transaction volume,average price,and increase/decrease are used as the input variables of the model,and LSTM is used The long and short-term memory network completes the preliminary prediction of the closing price,and uses the remaining three stock index historical price indicators such as the highest price,the opening price,and the lowest price to test the stability of the model,and obtain the output result of the LSTM neural network;Feature extraction of news text,using news feature vectors to correct the deviation of the initial LSTM network prediction results,and using the least square method to calculate the results to make the prediction results more accurate and create higher investment returns.The results show that the LSTM stock index price prediction fitting correction model,which incorporates news text features,has a satisfactory forecasting effect,and it has a good forecasting effect on the four price data sets.On the whole,the correction effect of the training set can be improved by nearly 30% compared with the preliminary prediction result,and the optimization of the test set can reach at least 10% or more,and the comparison with the prediction result of the SVR regression model verifies the fusion of news text in this article The characteristic LSTM stock index price prediction model has better fitting effect and higher prediction accuracy.