Model beto_sentiment_analysis_es

A finetuned model for Sentiment analysis in Spanish

This model was trained using Amazon SageMaker and the new Hugging Face Deep Learning container, The base model is BETO which is a BERT-base model pre-trained on a spanish corpus. BETO is of size similar to a BERT-Base and was trained with the Whole Word Masking technique.

BETO Citation

Spanish Pre-Trained BERT Model and Evaluation Data

@inproceedings{CaneteCFP2020,
  title={Spanish Pre-Trained BERT Model and Evaluation Data},
  author={Cañete, José and Chaperon, Gabriel and Fuentes, Rodrigo and Ho, Jou-Hui and Kang, Hojin and Pérez, Jorge},
  booktitle={PML4DC at ICLR 2020},
  year={2020}
}

Dataset

The dataset is a collection of movie reviews in Spanish, about 50,000 reviews. The dataset is balanced and provides every review in english, in spanish and the label in both languages.

Sizes of datasets:

• Train dataset: 42,500
• Validation dataset: 3,750
• Test dataset: 3,750

Intended uses & limitations

This model is intented for Sentiment Analysis for spanish corpus and finetuned specially for movie reviews but it can be applied to other kind of reviews.

Hyperparameters

{
"epochs": "4",
"train_batch_size": "32",    
"eval_batch_size": "8",
"fp16": "true",
"learning_rate": "3e-05",
"model_name": "\"dccuchile/bert-base-spanish-wwm-uncased\"",
"sagemaker_container_log_level": "20",
"sagemaker_program": "\"train.py\"",
}

Evaluation results

• Accuracy = 0.9101333333333333

• F1 Score = 0.9088450094671354

• Precision = 0.9105691056910569

• Recall = 0.9071274298056156

Test results

Model in action

Usage for Sentiment Analysis

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("edumunozsala/beto_sentiment_analysis_es")
model = AutoModelForSequenceClassification.from_pretrained("edumunozsala/beto_sentiment_analysis_es")

text ="Se trata de una película interesante, con un solido argumento y un gran interpretación de su actor principal"

input_ids = torch.tensor(tokenizer.encode(text)).unsqueeze(0)
outputs = model(input_ids)
output = outputs.logits.argmax(1)

Created by Eduardo Muñoz/@edumunozsala