---
language: su
tags:
  - sundanese-bert-base-emotion-classifier
license: mit
widget:
  - text: "Punten ini akurat ga ya sieun ihh daerah aku masuk zona merah"
---

## Sundanese BERT Base Emotion Classifier

Sundanese BERT Base Emotion Classifier is an emotion-text-classification model based on the [BERT](https://arxiv.org/abs/1810.04805) model. The model was originally the pre-trained [Sundanese BERT Base Uncased](https://hf.co/luche/bert-base-sundanese-uncased) model trained by [`@luche`](https://hf.co/luche), which is then fine-tuned on the [Sundanese Twitter dataset](https://github.com/virgantara/sundanese-twitter-dataset), consisting of Sundanese tweets.

10% of the dataset is kept for evaluation purposes. After training, the model achieved an evaluation accuracy of 96.82% and F1-macro of 96.75%.

Hugging Face's `Trainer` class from the [Transformers](https://huggingface.co/transformers) library was used to train the model. PyTorch was used as the backend framework during training, but the model remains compatible with other frameworks nonetheless.

## Model

| Model                                    | #params | Arch.     | Training/Validation data (text) |
| ---------------------------------------- | ------- | --------- | ------------------------------- |
| `sundanese-bert-base-emotion-classifier` | 110M    | BERT Base | Sundanese Twitter dataset       |

## Evaluation Results

The model was trained for 10 epochs and the best model was loaded at the end.

| Epoch | Training Loss | Validation Loss | Accuracy | F1       | Precision | Recall   |
| ----- | ------------- | --------------- | -------- | -------- | --------- | -------- |
| 1     | 0.759800      | 0.263913        | 0.924603 | 0.925042 | 0.928426  | 0.926130 |
| 2     | 0.213100      | 0.456022        | 0.908730 | 0.906732 | 0.924141  | 0.907846 |
| 3     | 0.091900      | 0.204323        | 0.956349 | 0.955896 | 0.956226  | 0.956248 |
| 4     | 0.043800      | 0.219143        | 0.956349 | 0.955705 | 0.955848  | 0.956392 |
| 5     | 0.013700      | 0.247289        | 0.960317 | 0.959734 | 0.959477  | 0.960782 |
| 6     | 0.004800      | 0.286636        | 0.956349 | 0.955540 | 0.956519  | 0.956615 |
| 7     | 0.000200      | 0.243408        | 0.960317 | 0.959085 | 0.959145  | 0.959310 |
| 8     | 0.001500      | 0.232138        | 0.960317 | 0.959451 | 0.959427  | 0.959997 |
| 9     | 0.000100      | 0.215523        | 0.968254 | 0.967556 | 0.967192  | 0.968330 |
| 10    | 0.000100      | 0.216533        | 0.968254 | 0.967556 | 0.967192  | 0.968330 |

## How to Use

### As Text Classifier

```python
from transformers import pipeline

pretrained_name = "sundanese-bert-base-emotion-classifier"

nlp = pipeline(
    "sentiment-analysis",
    model=pretrained_name,
    tokenizer=pretrained_name
)

nlp("Punten ini akurat ga ya sieun ihh daerah aku masuk zona merah")
```

## Disclaimer

Do consider the biases which come from both the pre-trained BERT model and the Sundanese Twitter dataset that may be carried over into the results of this model.

## Author

Sundanese BERT Base Emotion Classifier was trained and evaluated by [Wilson Wongso](https://w11wo.github.io/). All computation and development are done on Google Colaboratory using their free GPU access.

## Citation Information

```bib
@article{rs-907893,
    author   = {Wongso, Wilson
                and Lucky, Henry
                and Suhartono, Derwin},
    journal  = {Journal of Big Data},
    year     = {2022},
    month    = {Feb},
    day      = {26},
    abstract = {The Sundanese language has over 32 million speakers worldwide, but the language has reaped little to no benefits from the recent advances in natural language understanding. Like other low-resource languages, the only alternative is to fine-tune existing multilingual models. In this paper, we pre-trained three monolingual Transformer-based language models on Sundanese data. When evaluated on a downstream text classification task, we found that most of our monolingual models outperformed larger multilingual models despite the smaller overall pre-training data. In the subsequent analyses, our models benefited strongly from the Sundanese pre-training corpus size and do not exhibit socially biased behavior. We released our models for other researchers and practitioners to use.},
    issn     = {2693-5015},
    doi      = {10.21203/rs.3.rs-907893/v1},
    url      = {https://doi.org/10.21203/rs.3.rs-907893/v1}
}
```