Crypto Currency Price Prediction through Tweets Using NLP

Abstract

This research paper presents a comprehensive approach to sentiment analysis and stock price prediction using tweets from cryptocurrency influencers. The methodology involves several stages: environment setup, data loading, preprocessing, sentiment analysis using multiple machine learning models, and stock price prediction. The sentiment analysis leverages pre-trained models such as Roberta, VADER, XLNet, ALBERT, BERT, and BERTweet, each fine-tuned for aspect-based sentiment analysis. A majority voting mechanism determines the final sentiment, mitigating biases and enhancing reliability. Historical Bitcoin prices are fetched to analyze the relationship between tweet sentiments and price movements. The stock price prediction framework utilizes deep learning architectures, including LSTM, GRU, CNN-LSTM, Transformer, Bidirectional LSTM, and Simple CNN models. These models are trained on historical stock data, with features scaled and split into training, validation, and test sets. The models are evaluated using metrics such as RMSE, MSE, MAE, and R-squared, and their performance is visualized through plots of actual versus predicted prices and trends. The integration of sentiment analysis with stock price prediction provides a robust tool for understanding the impact of social media on financial markets. Using multiple models and majority voting ensures a comprehensive and reliable sentiment analysis, while the diverse deep-learning models offer insights into stock price trends and predictions. This research contributes to financial analytics by demonstrating the potential of combining NLP and deep learning techniques for market prediction.