|
|
--- |
|
|
license: mit |
|
|
datasets: |
|
|
- Superar/Puntuguese |
|
|
language: |
|
|
- pt |
|
|
base_model: |
|
|
- neuralmind/bert-base-portuguese-cased |
|
|
pipeline_tag: text-classification |
|
|
tags: |
|
|
- humor |
|
|
- pun |
|
|
- pun-recognition |
|
|
--- |
|
|
|
|
|
# Pun Recognition in Portuguese |
|
|
|
|
|
This is a Pun Recognition model for texts in Portuguese, as reported in two of our publications: |
|
|
|
|
|
- **Exploring Multimodal Models for Humor Recognition in Portuguese** ([PROPOR 2024 Paper](https://aclanthology.org/2024.propor-1.62/)) |
|
|
- **Puntuguese: A Corpus of Puns in Portuguese with Micro-Edits** ([LREC-COLING 2024 Paper](https://aclanthology.org/2024.lrec-main.1167/)) |
|
|
|
|
|
The model has been fine-tuned on the [Puntuguese](https://huggingface.co/datasets/Superar/Puntuguese) dataset, a collection of puns and corresponding non-pun texts in Portuguese. |
|
|
|
|
|
With this model, we achieved a maximum of **69% F1-Score** in the task of Pun Recognition with Puntuguese. |
|
|
|
|
|
## Installation and Setup |
|
|
|
|
|
To use this model, ensure you have the following dependencies installed: |
|
|
```bash |
|
|
pip install accelerate datasets scikit-learn torch transformers |
|
|
``` |
|
|
|
|
|
## How to Use |
|
|
To load the Puntuguese corpus and use the model for pun classification, run the following script: |
|
|
|
|
|
```python |
|
|
from datasets import load_dataset |
|
|
from transformers import pipeline |
|
|
import pandas as pd |
|
|
from sklearn.metrics import classification_report |
|
|
|
|
|
dataset = load_dataset('Superar/Puntuguese') |
|
|
classifier = pipeline('text-classification', model='Superar/pun-recognition-pt', device=0) |
|
|
|
|
|
prediction = classifier(dataset['test']['text']) |
|
|
pred_df = pd.DataFrame(prediction) |
|
|
pred_df['label'] = pred_df['label'].str[-1].astype(int) |
|
|
|
|
|
y_true = dataset['test']['label'] |
|
|
y_pred = pred_df['label'] |
|
|
print(classification_report(y_true, y_pred)) |
|
|
``` |
|
|
|
|
|
## Hyperparameters |
|
|
|
|
|
We used [Weights and Biases](https://wandb.ai/) to do a random search to optimize for the lowest evaluation loss using the following configuration: |
|
|
|
|
|
```python |
|
|
{ |
|
|
'method': 'random', |
|
|
'metric': {'name': 'loss', 'goal': 'minimize'}, |
|
|
'parameters': { |
|
|
'optim': {'values': ['adamw_torch', 'sgd']}, |
|
|
'learning_rate': {'distribution': 'uniform', 'min': 1e-6, 'max': 1e-4}, |
|
|
'per_device_train_batch_size': {'values': [16, 32, 64, 128]}, |
|
|
'num_train_epochs': {'distribution': 'uniform', 'min': 1, 'max': 5} |
|
|
} |
|
|
} |
|
|
``` |
|
|
|
|
|
The best hyperparameters found were: |
|
|
|
|
|
- **Learning Rate**: 8.47e-5 |
|
|
- **Optimizer**: AdamW |
|
|
- **Training Batch Size**: 128 |
|
|
- **Epochs**: 2 |
|
|
|
|
|
## Citation |
|
|
|
|
|
```bibtex |
|
|
@inproceedings{InacioEtAl2024, |
|
|
title = {Puntuguese: A Corpus of Puns in {{Portuguese}} with Micro-Edits}, |
|
|
booktitle = {Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation ({{LREC-COLING}} 2024)}, |
|
|
author = {In{\'a}cio, Marcio Lima and {Wick-Pedro}, Gabriela and Ramisch, Renata and Esp{\'{\i}}rito Santo, Lu{\'{\i}}s and Chacon, Xiomara S. Q. and Santos, Roney and Sousa, Rog{\'e}rio and Anchi{\^e}ta, Rafael and Goncalo Oliveira, Hugo}, |
|
|
editor = {Calzolari, Nicoletta and Kan, Min-Yen and Hoste, Veronique and Lenci, Alessandro and Sakti, Sakriani and Xue, Nianwen}, |
|
|
year = {2024}, |
|
|
month = may, |
|
|
pages = {13332--13343}, |
|
|
publisher = {{ELRA and ICCL}}, |
|
|
address = {Torino, Italia}, |
|
|
url = {https://aclanthology.org/2024.lrec-main.1167} |
|
|
} |
|
|
``` |
