metadata
license: mit
Model
Multi-language sentiment classification model developed over the Microsoft DeBERTa-v3 base model. In order to train the model the following dataset where used:
- tyqiangz/multilingual-sentiments
- cardiffnlp/tweet_sentiment_multilingual
- mteb/tweet_sentiment_multilingual
- Sp1786/multiclass-sentiment-analysis-dataset
- ABSC amazon review
- SST2
Evaluation and comparison with GPT-4o model:
| Dataset | Model | F1 | Accuracy |
|---|---|---|---|
| sst2 | Our | 0.6161 | 0.9231 |
| GPT-4 | 0.6113 | 0.8605 | |
| sent-eng | Our | 0.6289 | 0.6470 |
| GPT-4 | 0.4611 | 0.5870 | |
| sent-twi | Our | 0.3368 | 0.3488 |
| GPT-4 | 0.5049 | 0.5385 | |
| mixed | Our | 0.5644 | 0.7786 |
| GPT-4 | 0.5336 | 0.6863 | |
| absc-laptop | Our | 0.5513 | 0.6682 |
| GPT-4 | 0.6679 | 0.7642 | |
| absc-rest | Our | 0.6149 | 0.7726 |
| GPT-4 | 0.7057 | 0.8385 | |
| stanford | Our | 0.8352 | 0.8353 |
| GPT-4 | 0.8045 | 0.8032 | |
| amazon-var | Our | 0.6432 | 0.9647 |
| GPT-4 | 0.0000 | 0.9450 |
Reference
TBA