mrezadit/indobert-emotion-classification-v2

This model is a fine-tuned version of indobenchmark/indobert-base-p2. It is specifically designed to classify emotions in Indonesian text into 5 categories: Joy, Anger, Sadness, Fear, and Love. The model achieved an accuracy of 81.78% on the evaluation set.

Intended uses & limitations

• Use cases: Sentiment analysis, social media monitoring, and enhancing chatbot emotional intelligence.
• Limitations: Limitations: The model's performance may decrease when handling complex sarcasm or non-standard informal language that is less commonly used.

Training and evaluation data

The model was trained on 6000+ data samples, combining the Indonesian Twitter Emotion Dataset by meisaputri21, Dataset IndoNLU EmoT, and 1,000 generated data samples to improve robustness.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: AdamW with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
0.6961	1.0	320	0.6332	0.7709	0.7707
0.4698	2.0	640	0.5622	0.8092	0.8097
0.189	3.0	960	0.7534	0.7991	0.7993
0.1074	4.0	1280	0.9063	0.8100	0.8123
0.0548	5.0	1600	1.1038	0.8108	0.8130
0.0235	6.0	1920	1.1756	0.8092	0.8113
0.0167	7.0	2240	1.1835	0.8170	0.8182
0.0004	8.0	2560	1.1934	0.8202	0.8217
0.0003	9.0	2880	1.2152	0.8194	0.8207
0.0007	10.0	3200	1.2226	0.8178	0.8196

Framework versions

• Transformers 4.57.6
• Pytorch 2.9.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.2

Usage

from transformers import pipeline

# Load the model
classifier = pipeline("text-classification", model="mrezadit/indobert-emotion-classification-v2")
# If score < 0.85, classify as 'neutral'
threshold = 0.85

text = "Sumpah gue seneng banget hari ini!"
result = classifier(text)[0]

label = result['label'] if result['score'] >= threshold else "neutral"
print(f"Teks: {text}\nHasil: {label}")