Stock Trend Prediction Based On Time Series Diagram Neural Network And Investor Sentiment Analysis

Stock trend forecasting is the basic need for investors to obtain excess returns.The academic and industrial circles have never stopped exploring it.Many methods have been developed to try to predict the future trend of stocks.However,the real stock market is very complex and time-varying,which makes it extremely difficult to accurately predict it by traditional methods.With the breakthrough progress of deep learning,the Long Short-Term Memory Neural Networks(LSTM),which specializes in solving time series problems,has become one of the commonly used tools for stock forecasting.However,LSTM model is prone to weak generalization.In order to explore a more effective prediction model,this paper proposes a model of Temporal Graph Attention Network(TGAT)followed by Fully Connected Network to forecast the future trend of stocks,which can better capture the timing and dynamics of stock data.Affected by the COVID-19,pharmaceutical stocks have attracted much attention.This article selects the chemical and pharmaceutical sector index of the Eastmoney platform as the research object,and builds the investor sentiment features based on the stock comment text data,which is modeled and analyzed together with the technical indicators of the stock market.By comparing the loss function changes on the train set and test set of the model proposed in this paper and the LSTM model,it is shown that the model proposed in this paper can improve the weak generalization phenomenon of the model and better capture the dynamic characteristics of time series data.At the same time,the experimental results also show that the addition of investor sentiment features can improve the accuracy of stock trend prediction by 5 percentage points.