Comparative Sentiment Analysis of Cryptocurrency Apps Using BERT, RoBERTa, and ChatGPT

Abstract

Natural Language Processing (NLP) has greatly improved the ability to analyze and understand textual data, which is crucial for understanding user attitudes. Evaluating reviews from top bitcoin apps in India and Twitter data, this study applies advanced natural language processing techniques including BERT, RoBERTa, and the ChatGPT model to determine user sentiment. Using a diverse dataset consisting of 7,197 reviews, the author compared the models’ performance using metrics such as accuracy, precision, recall, and F1-score. According to the results, RoBERTa achieved the highest accurac y and F1 score (90%), followed by BERT (89.37% accuracy, 89% F1 score) and ChatGPT (90.00% accuracy, 88% F1 score). The performance metrics of conventional models, including Naive Bayes and Support Vector Machine (SVM), were poorer, showing that advanced natural language processing (NLP) models handled sentiment analysis better. Comparison bar charts and a confusio n matrix are two visualizations that help to further explain the findings. This research has real-world implications for understanding how people feel about bitcoin apps by revealing commonalities and missing features. To further enhance performance and accuracy, future research will optimize model parameters and explore sentiment analysis on a broader range of platforms.