Sentiment Based LSTM for Stock Price Prediction: Congress vs General Public

Stock market prediction has been challenging researchers for decades; nevertheless, new technology and methods becoming available are making a case that it could be done. One of the statistical approaches to stock market prediction is Time Series Forecasting. There are different Neural Networks designed for Time Series Forecasting; however, one form of neural network has been becoming more and more popular, Long Short-term Memory. The purpose of this paper is to predict accurately the change in stock price over the time-period of January 1, 2018-February 15, 2022. The technical data was collected from Yahoo Finance and include the stock information of the FAANG companies. Other technical indicators calculated are the Stochastic Oscillator Index (%K), William Index (%R), and relative strength index (RSI). This project will utilize the sentiment of elected congress officials, general public (from Twitter), and a combination of the two as inputs and test the prediction accuracy of the overall model. The LSTM was created with two methods, the first being SQL Server Data Tools, and the second being an implementation of LSTM using the Keras library. These results were then evaluated using accuracy, precision, recall, f-1 score, mean absolute error (MAE), root mean squared error (RMSE), and symmetric mean absolute percentage error (SMAPE). The results of this project were that the sentiment models all outperformed the control LSTM. The public model for Facebook on SQL Server Data Tools performed the best overall with 0.9743 accuracy and 0.9940 precision