Bitcoin Tweets Sentiment Classification Using an Enhanced BERT Model: A Transformer-Based Approach

Bitcoin’s price fluctuations have made it one of the most popular cryptocurrencies. However, scientists are beginning to realize how useful it is to use tweets to anticipate public opinion on a variety of topics, including political unrest and the economy. Tweets about cryptocurrencies often reflect the overall sentiment toward Bitcoin, either directly or indirectly. In this study, we used a dataset from the Data World platform to analyze Bitcoin-related tweets. We have prepared the dataset using standard preprocessing steps such as tokenization, removal of stop words, and text normalization. We employed various transformer-based models like Electra, RoBERTa, XLNet, and BERT base for sentiment categorization. Additionally, we have proposed an enhanced BERT base model, to predict the sentiment of Bitcoin-related tweets. These models analyze the tone of tweets to anticipate potential trends in Bitcoin behavior. Our enhanced BERT model achieved the highest accuracy of 98.4%, outperforming other transformer models, particularly the BERT base model. This superior performance is attributed to its robust encoding capabilities and the incorporation of an additional block within the model. This trained model can be used for the classification of Bitcoins-related tweets and the future trend can be determined based on the people’s sentiments about the Bitcoin.