Sentiment-driven forecasting: enhancing cryptocurrency price prediction with biLSTM and DistilBERT

Cryptocurrency is a secure and decentralised virtual currency that is digitally encrypted to prevent counterfeiting or double spending. Widely embraced as a popular investment, it is actively traded on blockchain-based crypto exchanges with continuously increasing transaction volumes. Cryptocurrency price prediction has been evolving from traditional time-series and statistical models to more sophisticated approaches, incorporating machine learning techniques. Early notable experts in the field had delved into applying machine learning for predicting prices of various cryptocurrencies like Bitcoin. The advent of LSTM and biLSTM architectures signalled a transition towards utilisation of Deep Learning for more precise predictions. In various studies, Deep Learning is employed to assess sentiments and emotions in texts. The proposed models such as LSTM-GRU model exhibit superior accuracy, surpassing both traditional machine learning models and advanced models. Despite of rapid progress in applications of these models to assess cryptocurrencies trends, a notable gap exists in incorporating sentiment analysis combined with prediction models which is crucial for understanding market dynamics. This study addresses the gap by combining biLSTM for price prediction with sentiment analysis using DistilBERT and Vader. This integration enhances predictive capabilities, capturing market sentiment nuances and improving cryptocurrency price forecasting accuracy. By infusing sentiment analysis into the biLSTM framework, this research presents a holistic model which considers historical price patterns and prevailing market sentiment.