Sentiment Analysis of Cryptocurrency Data: BERT vs. GPT-۲ - A Comparative Study

This paper compares the two language models, BERT and GPT-۲, in order to investigate sentiment in the context of cryptocurrency text data. A dataset of cryptocurrency-related text from various sources is labeled using a zero-shot classifier, and the BERT and GPT-۲ models are fine-tuned on this dataset. The study evaluates the performance of BERT and GPT-۲ in sentiment classification tasks, considering metrics such as recall, accuracy, and F۱-score. Results reveal that BERT performs more effectively than GPT-۲ in the area of understanding and classifying sentiment, specifically negative sentiment. The study emphasizes the significance of model goals and design for achieving superior performance in natural language processing tasks.

Keywords:

sentiment analysis , BERT , GPT-۲ , cryptocurrency