Deverta V3 Japanese Large Joy

🚀 DeBERTa Emotion Predictor

This package provides a DeBERTa-based model for predicting emotions in Japanese text. It uses fine-tuned DeBERTa models to estimate emotions in Japanese texts. You can easily obtain the predicted emotion labels and the confidence of the positive class for each text using models for eight emotions (Joy, Sadness, Anticipation, Surprise, Anger, Fear, Disgust, Trust).

🚀 Quick Start

This package simplifies the process of emotion prediction in Japanese text. With the pre - trained DeBERTa models, you can quickly get emotion prediction results for your text.

✨ Features

• Eight - emotion Estimation: Utilize fine - tuned models for each emotion to perform emotion estimation on texts.
• Flexible Input Formats: Accept single texts, text lists, or pandas Series as input and return results in DataFrame format.
• Efficient Inference: Designed to load models to the GPU only when necessary to reduce GPU memory usage.

📦 Installation

Use pip to install the package:

pip install deberta-emotion-predictor

Note that torch, transformers, and pandas are also required for the operation:

pip install torch 
pip install transformers
pip install pandas 

If you want to use a GPU, you need to install the NVIDIA GPU driver. Please refer to other materials for this.

💻 Usage Examples

Basic Usage

from deberta_emotion_predictor import DeBERTaEmotionPredictor
predictor = DeBERTaEmotionPredictor()
result = predictor.predict_emotions("今日はとても嬉しい！")
predictor.show_emotions(result)

Advanced Usage

Passing a list of texts

sample_texts = [
    "そうだ 京都、行こう。",
    "がんばるひとの、がんばらない時間。"
]
result_df = predictor.predict_emotions(sample_texts)
predictor.show_emotions(result_df)

Passing a single text

result_single = predictor.predict_emotions("新しい朝が来た。")
print(result_single)

Output DataFrame

The output DataFrame contains eight columns representing the presence of each emotion and the probability values of each emotion.

print(result_df)

📚 Documentation

Directory Structure

deberta_emotion_predictor/         
├── README.md                      # This description file
├── deberta_emotion_predictor.py   # Implementation of the DeBERTaEmotionPredictor class
│   └── tokenizer_DeBERTa_v3_large/ # Tokenizer
├── setup.py
├── pyproject.toml
├── README.md
├── LICENSE
└── usage.py                       

Required Environment

• Python 3.6 or higher
• PyTorch
• transformers
• pandas

📄 License

This project is licensed under the Creative Commons Attribution - ShareAlike 4.0 International (CC BY - SA 4.0) License.

Copyright (c) 2025 Yoichi Takenaka

This work is licensed under the Creative Commons Attribution - ShareAlike 4.0 International License. To view a copy of this license, visit https://creativecommons.org/licenses/by - sa/4.0/.

This project is based on:

• DeBERTa (https://huggingface.co/microsoft/deberta - v3 - large), licensed under the MIT License.
• DeBERTa Japanese Model (https://huggingface.co/globis - university/deberta - v3 - japanese - large), licensed under the CC BY - SA 4.0 License.

Any modifications or derivative works must also be distributed under the same CC BY - SA 4.0 License.

⚠️ Important Note

Since eight types of DeBERTa models are downloaded from Hugging - face, the first startup takes a long time. Subsequent executions will be faster.