Bitcoin Price Prediction Using a Deep Learning-based Hybrid Model with Sentiment Analysis and Attention Mechanism

Cryptocurrency, particularly Bitcoin, holds significant importance for investors and researchers due to its volatile price dynamics, which are influenced by various internal and external factors. The non-linear nature of cryptocurrency price fluctuations presents a considerable challenge in accurately predicting Bitcoin price changes. In addition to internal price fluctuations in the Bitcoin market, external public sentiment can also have a significant impact on Bitcoin prices. Based on the current situation, aimed to achieve more accurate prediction result of Bitcoin price, a hybrid forecasting model combining multiple high-precision technologies is proposed in this study. Bidirectional Long Short-Term Memory (Bi-LSTM) and Gated Recurring Unit (GRU) hybrid deep learning architectures are integrated in this model to analyse historical data, followed by the application of the FbProphet model to account for seasonal and temporal effects. To enhance prediction reliability, sentiment analysis of Bitcoin-related news is conducted using a combined Bidirectional Encoder Representations from Transformers (BERT) and Valence Aware Dictionary and sEntiment Reasoner (VADER) model, capturing the influence of public sentiment on price trends. Furthermore, an attention mechanism is incorporated to enable dynamic adjustment of the fusion weights, improving the model's adaptability to changing market conditions. Experimental results demonstrate the effectiveness of this hybrid model in producing more accurate Bitcoin price predictions. Future research will extend this model to other cryptocurrencies, further validating its robustness and exploring additional external factors affecting market prices.