Indobert-emotion-classification

Description

This model is a fine-tuned version of indobenchmark/indobert-base-p2. It is specifically designed to classify Indonesian text into 12 distinct emotion categories, capturing various nuances in local and informal communication.

Dataset Information

The model was trained on a programmatically generated Indonesian dataset curated to cover a wide range of emotional expressions.

• Total Data: 30,000 rows.
• Methodology: Scripted Data Generation with custom slang handling (mager, gabut, baper, goks, etc).

12-Emotion Categories:

Normal (Normal), Frustrasi (Frustrated), Jengkel (Annoyed), Marah (Angry), Lelah (Tired), Sedih (Sad), Sabar (Patient), Senang (Happy), Takut (Afraid), Terkejut (Surprised), Gila (Crazy), Cinta (Love)

Limitations

• Synthetic Data Bias: Since the model was trained on programmatically generated (synthetic) data, it may exhibit rigid patterns compared to natural human-to-human conversations.
• Slang Evolution: While the model supports current Indonesian slang, the rapid evolution of internet linguistics means that newer slang terms emerging after the training period may not be recognized accurately.
• Context Sensitivity: The model is optimized for short-form text classification. Accuracy may decrease for very long documents or texts containing complex, mixed emotions.
• Overfitting Risk: The extremely low loss values (approaching zero) suggest that the model is highly specialized to the specific patterns of this dataset, which might affect generalization on completely out-of-distribution data.

Training Results

Metric	Value
Epochs	10
Final Training Loss	0.000000
Final Validation Loss	0.000006
Optimization	AdamW

Usage

from transformers import pipeline
classifier = pipeline("text-classification", model="mrezadit/indobert-emotion-classification")
results = classifier("jujur gue gabut parah hari ini")
print(results)