SiEBERT: Sentiment-RoBERTa-Large-English
| Property | Value |
|---|---|
| Base Architecture | RoBERTa-large |
| Task | Binary Sentiment Analysis |
| Paper | Hartmann et al. 2023 |
| Downloads | 52,046 |
| Average Accuracy | 93.2% |
What is sentiment-roberta-large-english?
SiEBERT is a sophisticated sentiment analysis model built on RoBERTa-large architecture, specifically designed for binary sentiment classification in English text. Fine-tuned on 15 diverse datasets, it demonstrates exceptional generalization capabilities across various text types, from product reviews to social media content.
Implementation Details
The model employs a fine-tuned version of RoBERTa-large with carefully selected hyperparameters (learning rate: 2e-5, epochs: 3.0, warmup steps: 500). It's optimized for binary classification, outputting either positive (1) or negative (0) sentiment predictions.
- Outperforms standard DistilBERT SST-2 model by 15 percentage points
- Achieves 93.2% average accuracy across diverse datasets
- Implements efficient weight decay of 0.01
- Supports both direct inference and transfer learning
Core Capabilities
- Binary sentiment classification for various text types
- Robust performance across multiple domains (reviews, tweets, etc.)
- Easy integration with Hugging Face pipelines
- Suitable for transfer learning and further fine-tuning
Frequently Asked Questions
Q: What makes this model unique?
SiEBERT's uniqueness lies in its comprehensive training across 15 different datasets, resulting in superior generalization capability compared to models trained on single datasets. It maintains high performance even in leave-one-out evaluations, demonstrating robust cross-domain adaptation.
Q: What are the recommended use cases?
The model is ideal for analyzing sentiment in product reviews, social media posts, customer feedback, and general text analysis. It's particularly effective when dealing with diverse text sources or when cross-domain generalization is required.
