Sentiment-based Bitcoin Movement Prediction with Deep Learning

Abstract

Bitcoin movement prediction has been a research topic, especially with the Bitcoin bubble period during the COVID-10 pandemic era. It has been discussed that sentiment factors are influential in the prediction process. In this study, to validate the helpfulness of these emotion-related features, three open datasets with sentiment are used, and three types of neural networks, including LSTM, GRU, and TCN, are chosen as the prediction models, using a regression model to study the relationship between them. The result shows that the sentiment-based factors did help in two of the three datasets.