RoBERTa-Sentimentic 🎭
Fine-tuned RoBERTa for sentiment analysis, trained on 50k+ samples from IMDB and Stanford SST-2.
Performance
- IMDB: 89.2% accuracy (+39.6% improvement)
- SST-2: 91.5% accuracy (domain-specific)
- Cross-domain: 87.7% accuracy (IMDB→SST)
Usage
from transformers import pipeline
classifier = pipeline("sentiment-analysis", model="abhilash88/roberta-sentimentic")
result = classifier("This movie is amazing!")
Training Details
- Base Model: roberta-base
- Datasets: IMDB (25k) + Stanford SST-2 (25k)
- Method: Domain-specific fine-tuning with class balancing
- Performance: Significant improvements across both datasets
Results Summary
| Model | IMDB | SST-2 | Notes |
|---|---|---|---|
| Pre-trained | 49.5% | 49.1% | Baseline |
| Fine-tuned | 89.2% | 91.5% | Optimized |