A Multimodal Model with Twitter Finbert Embeddings for Extreme Price Movement Prediction of Bitcoin

Bitcoin, with its ever-growing popularity, has demonstrated extreme price volatility since origin. This volatility, together decentralised nature, make Bitcoin highly subjective to speculative trading as compared more traditional assets. In this paper, we propose a multimodal model for predicting fluctuations. takes input variety of correlated assets, technical indicators, well Twitter content. an in-depth study, explore whether social media discussions from the general public on have predictive power movements. A dataset 5,000 tweets per day containing keyword `Bitcoin' was collected 2015 2021. dataset, called PreBit, is made available online. our hybrid model, use sentence-level FinBERT embeddings, pretrained financial lexicons, so capture full contents and feed it in understandable way. By combining these embeddings Convolutional Neural Network, built significant market The final ensemble includes NLP based candlestick data, indicators asset prices. ablation contribution individual modalities. Finally, backtest strategy predictions models varying prediction threshold show that can used build profitable reduced risk over `hold' or moving average strategy.