Forecasting the Early Market Movement in Bitcoin Using Twitter's Sentiment Analysis: An Ensemble-based Prediction Model

Ahmed Ibrahim

DOI: https://doi.org/10.1109/iemtronics52119.2021.9422647

2021-04-21

Abstract:Data collected from social media such as tweets, posts, and blogs can assist in an early indication of market sentiment in the financial field. This has frequently been conducted on Twitter data in particular. Using data mining techniques, opinion mining, machine learning, natural language processing (NLP), and knowledge management, the underlying public mood states and sentiment can be uncovered. As cryptocurrencies play an increasingly significant role in global economies, there is an evident relationship between Twitter sentiment and future price fluctuations in Bitcoin. This paper assesses Tweets' collection, manipulation, and interpretation to predict early market movements of cryptocurrency. More specifically, sentiment analysis and text mining methods, including Logistic Regressions, Binary Classified Vector Prediction, Support Vector Mechanism, and Naive Bayes, were considered. Each model was evaluated on their ability to predict public mood states as measured by ‘tweets' from Twitter during the era of covid-19. An XGBoost-Composite ensemble model is constructed, which achieved higher performance than the state-of-the-art prediction models.