roberta-base-stocktwits-finetuned
| Property | Value |
|---|---|
| Base Architecture | RoBERTa-base |
| Training Data | 3.2M StockTwits comments |
| Best Validation Accuracy | 93.43% |
| Model URL | huggingface.co/zhayunduo/roberta-base-stocktwits-finetuned |
What is roberta-base-stocktwits-finetuned?
This is a specialized sentiment analysis model developed by NUS ISS students, fine-tuned on RoBERTa-base architecture using 3.2 million StockTwits comments. The model is specifically designed to classify financial market sentiment into 'Bullish' or 'Bearish' categories, making it particularly valuable for analyzing stock market-related discussions and comments.
Implementation Details
The model was trained using a careful optimization strategy with a batch size of 32 and learning rate of 2e-5. Training progressed through 4 epochs, showing consistent improvement in validation accuracy from 86.79% to 93.43%. The implementation includes sophisticated text preprocessing, handling URLs, emojis, cashtags ($), hashtags (#), and user mentions.
- Advanced text preprocessing pipeline for social media content
- Optimized training parameters for financial sentiment analysis
- Support for both simple text and complex social media formats
Core Capabilities
- Binary sentiment classification (Bullish/Bearish)
- Processing of financial-specific tokens and symbols
- Handling of social media formatting and emoji conversion
- High accuracy on market sentiment detection
Frequently Asked Questions
Q: What makes this model unique?
The model's specialization in financial market sentiment and its training on a massive dataset of 3.2M StockTwits comments makes it particularly effective for analyzing stock market-related discussions. The high validation accuracy of 93.43% demonstrates its reliability for real-world applications.
Q: What are the recommended use cases?
The model is ideal for analyzing stock market comments, social media discussions about financial markets, and automated sentiment analysis of trading-related content. It's particularly effective with short-form content similar to StockTwits posts.
