|
|
--- |
|
|
license: mit |
|
|
tags: |
|
|
- indobert |
|
|
- emotion-classification |
|
|
- text-classification |
|
|
- indonesian |
|
|
- torch |
|
|
language: |
|
|
- id |
|
|
datasets: |
|
|
- PRDECT-ID |
|
|
model-index: |
|
|
- name: IndoBERT Emotion Classification (5-Class) |
|
|
results: |
|
|
- task: |
|
|
type: text-classification |
|
|
name: Emotion Classification |
|
|
dataset: |
|
|
name: PRDECT-ID |
|
|
type: text |
|
|
description: > |
|
|
A dataset of Indonesian product reviews labeled with five emotion |
|
|
categories: love, happiness, anger, fear, and sadness. |
|
|
metrics: |
|
|
- name: Accuracy |
|
|
type: accuracy |
|
|
value: 0.7167 |
|
|
- name: F1 Score |
|
|
type: f1 |
|
|
value: 0.7125 |
|
|
- name: Precision |
|
|
type: precision |
|
|
value: 0.7179 |
|
|
- name: Recall |
|
|
type: recall |
|
|
value: 0.7167 |
|
|
base_model: |
|
|
- indobenchmark/indobert-base-p1 |
|
|
--- |
|
|
|
|
|
# IndoBERT Emotion Classification (5-Class) |
|
|
|
|
|
Model ini merupakan hasil *fine-tuning* dari [`indobenchmark/indobert-base-p1`](https://huggingface.co/indobenchmark/indobert-base-p1) untuk tugas klasifikasi emosi dalam Bahasa Indonesia, dengan 5 label emosi: `love`, `happiness`, `anger`, `fear`, dan `sadness`. |
|
|
|
|
|
## 🧠 Dataset |
|
|
|
|
|
Model ini dilatih menggunakan **PRDECT-ID Dataset**, yaitu kumpulan ulasan produk berbahasa Indonesia dari e-commerce Tokopedia, yang sudah dianotasi dengan label emosi oleh ahli psikologi klinis. |
|
|
|
|
|
- 29 kategori produk |
|
|
- Anotasi emosi oleh tim profesional |
|
|
- Setiap entri memiliki 1 label emosi |
|
|
|
|
|
## 🛠 Fine-tuning Details |
|
|
|
|
|
- **Base model**: `indobenchmark/indobert-base-p1` |
|
|
- **Training epochs**: 5 dari total 10 (early stopping dengan `load_best_model_at_end=True`) |
|
|
- **Batch size**: 8 |
|
|
- **Learning rate**: 2e-5 |
|
|
- **Weight decay**: 0.05 |
|
|
- **Validation strategy**: per epoch |
|
|
- **Evaluation metric**: `eval_accuracy` (dengan `greater_is_better=True`) |
|
|
- **Cross-validation**: Stratified K-Fold (n_splits=5) |
|
|
|
|
|
### Eval Results (Best Model @ Epoch 3) |
|
|
|
|
|
| Metric | Value | |
|
|
|-------------|---------| |
|
|
| Accuracy | 0.7167 | |
|
|
| F1 Score | 0.7125 | |
|
|
| Precision | 0.7179 | |
|
|
| Recall | 0.7167 | |
|
|
| Eval Loss | 0.7614 | |
|
|
|
|
|
## 🚀 How to Use |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline |
|
|
|
|
|
model = AutoModelForSequenceClassification.from_pretrained("galennolan/indobert-b-p1-indoemotion-5class") |
|
|
tokenizer = AutoTokenizer.from_pretrained("galennolan/indobert-b-p1-indoemotion-5class") |
|
|
|
|
|
emotion_classifier = pipeline("text-classification", model=model, tokenizer=tokenizer) |
|
|
|
|
|
emotion_classifier("Produk ini bikin aku senang banget!") |
