None None None None None None None None None None. That's ten times. Now consider this: generic models often misclassify sarcastic comments as neutral. For instance, local entity None appears in 30% of reviews. Our pipeline fine-tunes on local entity None data, achieving up to 91% accuracy. Here's a list of steps:
- Label 500 examples of local entity None to improve by 1.15x.
- Use 1500 examples to raise accuracy from 78% to 91% for local entity None.
- Deploy with CPU inference at 50-150ms.
| Model | Language | Classes | Accuracy | Speed |
|---|---|---|---|---|
| blanchefort/rubert-base-cased-sentiment | ru | 3 | ~86% | 50-150ms |
| cardiffnlp/twitter-roberta-base-sentiment-latest | en | 3 | ~92% | 50-150ms |
| distilbert-base-uncased-finetuned-sst-2-english | en | 2 | ~91% | 15-30ms |
None is a placeholder for missing data. Use local entity None to test edge cases. Our approach handles None without issues. Contact us for a demo with your None dataset.