|
|
--- |
|
|
language: |
|
|
- es |
|
|
- en |
|
|
tags: |
|
|
- sentiment-analysis |
|
|
- xlm-roberta |
|
|
- multilingual |
|
|
- movies |
|
|
license: apache-2.0 |
|
|
base_model: |
|
|
- FacebookAI/xlm-roberta-base |
|
|
--- |
|
|
|
|
|
# XLM-R Sentiment EN/ES (Movie Reviews) |
|
|
|
|
|
Clasificador binario (*Positive/Negative*) para reseñas de películas en **inglés y español**, fine-tuned desde `xlm-roberta-base` con **Rotten Tomatoes movies and critic reviews dataset** from [Kaggle](https://www.kaggle.com/datasets/stefanoleone992/rotten-tomatoes-movies-and-critic-reviews-dataset) |
|
|
|
|
|
**Métricas:** |
|
|
|
|
|
Acc **0.8519** · F1 **0.8876** · Prec **0.8646** · Rec **0.9119** · AUC **0.9260** |
|
|
*Umbral recomendado:** **0.48* |
|
|
|
|
|
## Uso rápido |
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification |
|
|
import torch |
|
|
m = "Ricardouchub/xlmr-sentiment-es-en"; thr = 0.48 |
|
|
tok = AutoTokenizer.from_pretrained(m, use_fast=True) |
|
|
mdl = AutoModelForSequenceClassification.from_pretrained(m).eval() |
|
|
enc = tok(["Excelente actuación, final predecible."], truncation=True, max_length=224, padding=True, return_tensors="pt") |
|
|
p = torch.softmax(mdl(**enc).logits, dim=-1)[:,1].item() |
|
|
print(("POSITIVE" if p>=thr else "NEGATIVE"), round(p*100,1), "%") |
|
|
``` |
|
|
|
|
|
*Notas: split por película (evita fuga); limpieza mínima de texto. No apto para usos sensibles.* |
|
|
|
|
|
**Autor: Ricardo Urdaneta** |
