Enhancing Stock Price Prediction with Sentiment Analysis : A Comparison of LSTM Models

Abstract: Since the early 1980s stock traders and investment firms have been devloping prediction based machine learning models in an attempt to become rich. The problem is that stock price prediction is hard, and simply using historical price data is not enough to get accurate predictions. This thesis aims to determine if the use of social media sentiment in training a machine learning model for stock price prediction can improve the model’s accuracy. This was achieved through comparing two models, one trained solely on historical stock data, and one trained on historical stock data as well as public sentiment of the stock. Through sentiment analysis a large number of social media posts were ranked and added to the training dataset. The conclusion reached is that the prediction became slightly better as social media sentiment was added to the model parameters, thus making it an important addition to improving the tools for stock traders around the globe.