Clustering Social Media Data for Bitcoin Price Prediction with Transformer Model

Abstract

This paper explores the integration of social media data and natural language processing methods, specifically utilizing the Transformer model, to predict Bitcoin price movements. We aim to evaluate the effectiveness of using social media data and the Transformer model in forecasting market trends for Bitcoin. By analyzing social media posts and incorporating them into predictive models, we demonstrate the potential of the Transformer architecture in capturing complex dependencies and patterns within sequential Bitcoin prices. Additionally, different clustering methods are applied to process the original social media data in a rolling manner. The evaluation of Transformer-based models on historical data showcases their predictive performance compared to various social media data clustering approaches. Furthermore, the impact of incorporating outliers of social media data into the Transformer model is explored to improve prediction accuracy. The results of this study demonstrate the potential of clustering on social media data and the Transformer model for forecasting market trends.1