Predicting Bitcoin price movement through Sentiment Analysis: A Comprehensive Study

The people's feelings in tweets can affect on the direction of Bitcoin price. The study aimed to figure out how much Twitter posts affect the price movement of Bitcoin. It was tried to find accurate prediction models through a Sentiment Analysis (SA) that can predict these changes. Data collection was included real-time Twitter data from January 31, 2023, to June 6, 2023, focusing on English tweets containing the keyword "bitcoin". These data were paired with hourly Bitcoin price data, covering open, high, low, close, and volume values. Two pre-trained RoBERTa models (Tweetnlp and BERTweet) were used to perform SA on the collected tweets. Three neural network models (Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bidirectional LSTM (BiLSTM)) were explored and evaluated at different temporal granularities ranging from 2 hours to 17 hours. The findings showed that combining SA could improve prediction accuracy in a specific time horizon. The highest accuracy (90.3%) was achieved by a 2-layer GRU incorporating Tweetnlp sentiment analysis at a 16-hour lag. The research offers valuable insights into the role of SA in understanding and potentially predicting fluctuations in Bitcoin prices, highlighting its significance in the realm of cryptocurrency analysis. Future research could explore additional factors affecting the connection between social media sentiment and Bitcoin price.