Bitcoin Price Prediction Using Sentiment Analysis

In this study, we aim to enhance the capability of modern techniques used in predicting Bitcoin price movement. While the current methods can only predict the direction of price movement, our objective is to predict the magnitude of increase or decrease as well. To achieve this, we incorporate Twitter sentiment and tweet volume as variables. Through our experimentation, we seek to determine the best time window for which sentiment stated in tweets is a consistent pointer of future cost variation. We calculate the presentation of two different neural network models, a recurrent network-based and the other on Convolutional Neural Networks. The model we have developed aims to forecast the degree of change by formulating it as a multi-class classification task. We illustrate that this approach can result in more precise forecasts (up to 63.3%) when used in conjunction with a cost trend forecasting design. The primary contribution of our research is the ability to predict both the direction and magnitude of price change with a relatively high degree of accuracy. To accomplish our goals, we utilize Python as the primary programming language along with various libraries including Pandas, NumPy, matplotlib, numba, sklearn, bs4, pyspark, nltk, re, Pytorch, and light libraries. All our experiments are conducted in Jupyter notebooks and collect Tweets using GetOldTweets-python and augmento.