A STOCK PRICE PREDICTION MODEL BASED ON INVESTOR SENTIMENT AND OPTIMIZED DEEP LEARNING

Abstract

The research presents a model, MS-SSALSTM that combines multi-source data, sentiment analysis, swarm intelligence algorithms, and deep learning models to enhance stock price predictions. This model combines sentiment analysis based on postings on the East Money forum, creating a unique sentiment lexicon and calculating a sentiment index. This gives major information about the effect of market sentiment on the price of stocks. Sparrow Search Algorithm (SSA) is used to optimize the hyperparameters of LSTM, and it improves the accuracy of prediction. The efficacy of the MS-SSALSTM model is outstanding as proven by experiments. It is a priceless tool in accurate stock price prediction. The model is tailored to the volatile financial market in China, and is focused on forecasting stock prices in the short term, which can be used to make fast decisions by investors. A stock sentiment classification hybrid LSTM and GRU model was created. An ensemble strategy was adopted, considering a Voting Classifier (AdaBoost + RandomForest) to perform sentiment analysis and Voting Regressor (LinearRegression + RandomForestRegressor + KNeighborsRegressor) to predict the price of stock. They were seamlessly integrated with the models already (MLP, CNN, LSTM, MS-LSTM, MS-SSA-LSTM) and enhanced the overall predictive capabilities. An easy to use Flask structure that supports SQLite was developed to increase user interaction and testing and make the signup, signin, and model evaluation process easier.