Indonesian RoBERTa Base Emotion Classifier

Introduction

The Indonesian RoBERTa Base Emotion Classifier is a language model designed to classify emotions in Indonesian text. It is built on the Indo-RoBERTa model and has been fine-tuned using the IndoNLU EmoT dataset. The model achieves an F1-macro score of 72.05%, accuracy of 71.81%, precision of 72.47%, and recall of 71.94%.

Architecture

The classifier employs the Indo-RoBERTa architecture, a variant of the RoBERTa model optimized for the Indonesian language. It has been adapted specifically for emotion classification tasks, leveraging transfer learning techniques on the IndoNLU EmoT dataset.

Training

The model was trained across 7 epochs with a learning rate of 2e-5. Key performance metrics are reported per epoch, with the model achieving its best accuracy (72.5%) and F1 score (73.36%) in the final epoch.

Guide: Running Locally

To use the model for text classification:

from transformers import pipeline

pretrained_name = "StevenLimcorn/indonesian-roberta-base-emotion-classifier"
nlp = pipeline(
    "sentiment-analysis",
    model=pretrained_name,
    tokenizer=pretrained_name
)

result = nlp("Hal-hal baik akan datang.")

Suggested Cloud GPUs

The model was developed using Google Colaboratory, which offers free GPU access suitable for running this model.

License

The Indonesian RoBERTa Base Emotion Classifier is distributed under the MIT License, allowing for broad use and modification. Please ensure compliance with this license when using or distributing the model.