Transformer-Based Sentiment Classification of Blockchain and Cryptocurrency Tweets Using the DistilBERT Architecture

The rapid growth of blockchain technology and cryptocurrency adoption has significantly increased public discussions on social media platforms such as Twitter, where users frequently share opinions, investment perspectives, and market-related information. However, blockchain-related tweets are often characterized by noisy text, informal language, abbreviations, and cryptocurrency-specific terminology, making sentiment analysis tasks challenging for traditional natural language processing approaches. Therefore, this study proposes a transformer-based sentiment analysis framework using the DistilBERT model to classify blockchain-related tweets into Positive, Neutral, and Negative sentiment categories. The dataset initially consisted of 10,000 tweets collected from Twitter, and after preprocessing and data cleaning, 9,697 tweets were retained for analysis. Several preprocessing techniques were applied, including lowercase transformation, URL removal, mention removal, hashtag cleaning, punctuation removal, and whitespace normalization. Sentiment labeling was performed using the VADER sentiment analyzer. The experimental results demonstrate that the proposed DistilBERT model achieved strong classification performance with an accuracy of 93.61%, Macro Precision of 85.85%, Macro Recall of 85.35%, and Macro F1-score of 85.57%. The findings indicate that transformer-based approaches are highly effective for analyzing blockchain-related social media text and have strong potential for applications in cryptocurrency market monitoring and public opinion analysis.