Stock Market Prediction Based On LSTM And Investor Sentiment Analysis

The fluctuation of the stock price has a great influence on all aspects of social and economic life.Therefore,it is a hot topic for current researchers to predict the change trend of the stock price reasonably and effectively.The core of stock forecasting is to predict future stock market changes by learning from existing historical data.The traditional stock forecasting model adopts a benchmark model based on financial time series.However,the financial time series model ignores the influence of investor sentiment on stock market movements.In order to more accurately predict stock market changes,this paper establishes an LSTM-based model investor sentiment stock market forecasting model,the main work of this article contains the following two aspects:Firstly,an LSTM-based stock market forecasting model is established.This model uses LSTM to model the stock price time series directly,and on this basis,basic emotion features are incorporated to enhance model prediction performance.The LSTM model is one of the recursive neural network models.It can learn the laws of long-term dependence more effectively than the general neural network model.Therefore,based on the time series theory,this paper establishes a time series model based on LSTM.Compared with the classic model based on time series,this model has a certain improvement in the accuracy of prediction,and for the classical model,which ignores the influence of investor sentiment on the stock market changes.This paper extracts the emotional characteristics of the stock market texts using the text analysis method based on emotional dictionary.Emotional features in the text are combined with the time series of stock prices to construct a LSTM model that fuses basic emotion features.Compared with the LSTM-based time series model,this model has a significant improvement in accuracy.Secondly,on the basis of the above model,we use CNN to extract the deep emotional information to replace the basic sentiment features,and introduce additional information sources such as fundamental features on the data source level to further improve the model's prediction performance.Since the method of text analysis based on sentiment dictionary depends to a large extent on the quality and coverage of the dictionary itself,this paper uses CNN and pre-trained word vectors to conduct sentiment analysis on the text and combines the obtained features with stock time series.A stock prediction model that fuses deep emotional characteristics is constructed.After experiments,compared with the LSTM model that integrates the basic emotional characteristics,the stock prediction model that fuses the deep emotional characteristics has a great improvement in the accuracy of the prediction,which proves that CNN can more effectively extract the emotion in the text.And then improve the accuracy of model prediction.In terms of data sources,information sources of fundamental features were further added.Experiments confirm that the model accuracy has been further improved.This also certifies that multi-information sources can improve model prediction accuracy more effectively than single information sources.