Stock Price Prediction with Social Media Sentiment

Abstract

This thesis investigates the correlation effects between social media sentiments and the stock price of AMZN and TSLA, by utilizing pre-trained machine learning models, so-called transformers, and lexicon-based models. The comments were fetched from two sources, Reddit and Twitter. Moreover, two different approaches to incorporating the sentiment for stock price prediction were implemented. Firstly, moving average sentiment cross-over signals were studied and compared with the buy-and-hold strategy, as a baseline. Secondly, a Long Short-Term Memory neural network, with the sentiment as an additional feature, was implemented and compared to a classic Long Short-Term Memory network which only utilizes the previous stock prices as input for the prediction. The study showed evidence of significant correlation. The results indicate that social media sentiment can prove useful for stock market predictions and that there is a need for further and more extensive research on the topic in order to make more general claims. Furthermore, the transformer models turned out to not be superior to the lexicon-based model.