Date of Award

Spring 2025

Thesis Type

Rollins Access Only

Degree Name

Honors Bachelor of Arts

Department

Computer Science

Sponsor

Daniel Myers

Committee Member

JJ Jasser, Marc Sardy, Eunjung Noh

Abstract

This research investigates the relationship between social media sentiment and cryptocurrency price movements, with a focus on Bitcoin (BTC) and Ethereum (ETH). Utilizing state-of-the-art transformer models—BERT and RoBERTa—we analyze over 1.2 million social media posts collected from platforms such as Twitter, Reddit, Telegram, and Discord during 2022–2023. The results reveal statistically significant correlations between sentiment shifts and subsequent price movements, with higher correlations during bull markets (0.45–0.48) and lower during bear markets (0.22–0.28). We observe a typical 2–3 day lag between sentiment changes and price reactions. Sentiment classification achieved high accuracy rates (BERT: 85.1%, RoBERTa: 86.7%), outperforming simpler lexicon and time-series methods. The study further analyzes lag effects, volatility clustering, market regime behaviors, and cross-platform sentiment propagation. We demonstrate that social media sentiment serves not only as a reflection of market mood but also as a predictive signal for asset price dynamics. Practical implications include enhanced trading strategies, improved risk management, and tools for detecting market manipulation. Regulatory relevance is also highlighted through the development of sentiment-based monitoring tools to support market surveillance. Overall, the study contributes a robust deep learning framework for understanding and leveraging the influence of social sentiment in the rapidly evolving cryptocurrency landscape.

Download

Available for download on Wednesday, May 03, 2028

Share

COinS