Abstract
Many traders rely on Twitter tweets to guide their daily cryptocurrency trading decisions. This study explores the feasibility of automated sentiment analysis for cryptocurrencies, focusing on the altcoin NEO. Over five years of daily tweets containing NEO-related hashtags were collected, filtered, and manually labeled with positive, negative, or neutral sentiments. A Random Forest classifier achieved 77% accuracy in sentiment prediction. The research also investigated correlations between daily tweet sentiment and NEO price movements, revealing notable links between tweet volume, sentiment, and cryptocurrency price trends.
Introduction
The rise of digital payments and blockchain technology has fueled the growth of cryptocurrencies like Bitcoin (BTC) and Ethereum (ETH). These digital assets operate on decentralized networks, enabling secure peer-to-peer transactions without intermediaries.
Key Cryptocurrency Concepts:
- Blockchain: A decentralized ledger technology introduced by Satoshi Nakamoto.
- Altcoins: Alternative cryptocurrencies to Bitcoin (e.g., NEO, often called "China’s Ethereum").
- Crypto Exchanges: Platforms where cryptocurrencies are traded (e.g., Binance, Coinbase).
NEO aims to digitize physical assets using a three-tier system (smart contracts, digital assets, digital identities). Its price often correlates with BTC and ETH, but social media sentiment—particularly on Twitter—may also influence its market behavior.
Methodology
1. Data Collection
- Twitter Data: 3.1 million tweets with NEO-related hashtags (e.g.,
#NEO,$NEO) were collected using Python’s Twint library. - Price Data: Daily BTC, ETH, and NEO prices were sourced from Yahoo Finance.
2. Data Filtering
- Language: Retained only English tweets (846,790 tweets).
- Relevance: Manually filtered tweets mentioning NEO (44,124 tweets).
- Sentiment Labeling: A subset of 1,200 tweets (400 per sentiment) was manually tagged for training.
3. Sentiment Analysis
- Random Forest Classifier: Achieved 77% test accuracy using TF-IDF vectorization.
- BERT Model: For comparison, a pre-trained BERT model yielded 45% accuracy, highlighting the challenge of crypto-specific jargon.
4. Correlation Analysis
Examined relationships between:
- NEO price and BTC/ETH prices.
- Tweet sentiment and NEO price movements.
Key Findings
Sentiment Accuracy:
- Random Forest outperformed BERT, suggesting domain-specific training improves results.
- Neutral tweets showed the strongest correlation with NEO price (45%).
Price Correlations:
- NEO and ETH: 67% correlation.
- NEO and BTC: 41% correlation.
- BTC and ETH: 91% correlation.
Tweet Volume Impact:
- Higher tweet volumes often preceded price increases, indicating sentiment-driven market reactions.
Conclusion
This study demonstrates that automated sentiment analysis of Twitter data can predict cryptocurrency price trends with moderate accuracy. Key takeaways:
- Random Forest classifiers are effective for crypto-specific sentiment analysis.
- Neutral sentiment and tweet volume strongly correlate with NEO price movements.
- Future work could expand to other altcoins and refine models with larger datasets.
👉 Explore more about cryptocurrency trading strategies
FAQs
1. How accurate is Twitter sentiment for predicting crypto prices?
- In this study, 77% accuracy was achieved using a Random Forest model. Neutral tweets showed the strongest price correlation.
2. Why did BERT perform poorly compared to Random Forest?
- BERT lacked training on crypto-specific slang (e.g., "HODL," "whale"), reducing its effectiveness.
3. Which cryptocurrencies were analyzed?
- Focused on NEO, with comparative data from BTC and ETH.
4. What’s the significance of neutral tweets?
- Neutral tweets often reflect market stability and had the highest correlation (45%) with NEO price.
5. Can this method be applied to other altcoins?
- Yes, but altcoin-specific data collection and labeling are required for optimal results.
👉 Learn how to leverage sentiment analysis for crypto investments
This research bridges social media analytics and cryptocurrency trading, offering actionable insights for traders and developers. For full datasets and code, visit the project’s GitHub repository.