Crypto Sentiment Classifier - DistilRoBERTa
Model Description
This fine-tuned DistilRoBERTa model classifies financial/crypto tweets into BEARISH (0), BULLISH (1), or NEUTRAL (2) sentiments. Trained on 9.5k+ Twitter financial news samples with class weighting for imbalance.
Intended Uses
- Real-time sentiment analysis on crypto tweets.
- Integration into trading bots or dashboards.
Performance
| Metric | Value |
|---|---|
| Eval F1 (weighted) | 0.8811 |
| Accuracy | 0.8790 |
Limitations & Bias
- Heavy Neutral skew may underperform on extreme sentiments.
- Trained on general financial tweets; fine-tune further for pure crypto jargon.
- Downloads last month
- 707