CantoSent-Qwen-7B-OpenRice-v3

🐉 CantoSent-Qwen-7B-OpenRice-v3

📌 Model Summary

CantoSent-Qwen-7B-OpenRice-v3 is a fine-tuned version of stvlynn/Qwen-7B-Chat-Cantonese, designed for Cantonese sentiment classification on OpenRice restaurant reviews.
The task is a binary classification: positive (3–5 stars) vs negative (1–2 stars).
This model is optimized for research and academic use in low-resource language sentiment analysis, especially in the domain of Cantonese NLP.

---

🧠 Model Details

• Base Model: stvlynn/Qwen-7B-Chat-Cantonese
• Fine-tuned by: Geng Le
• Language: Cantonese (Chinese Traditional)
• Task: Binary Sentiment Classification
• Dataset: Cantonese OpenRice Reviews
• Emoji Handling: Translate emojis into text descriptions (with-emojis-texts version)

---

📊 Evaluation Results

Evaluated on a held-out test set (10% of 60,000 examples).

---

📁 Dataset Information

• Source: OpenRice review dataset from a public GitHub repository
• Size: 60,000 examples
• Labels:
  • Positive = 3–5 stars
  • Negative = 1–2 stars
• Split:
  • 90% training, 10% testing
• Emoji Handling: Version without emojis used in this training

---

🎯 Use Cases

• Cantonese sentiment analysis for food review systems
• Research in low-resource or regional dialect NLP
• Benchmarking multilingual/multi-dialect LLMs
• Educational or academic projects

---

🔬 Intended Use & Limitations

This model is research-priority. It is intended for:

• Academic research
• Model benchmarking
• Sentiment classification on food/dining-related Cantonese text

Limitations:

• Domain-specific (trained only on restaurant reviews)
• Does not support emojis in the current version
• Performance may degrade on informal or non-Cantonese text

---

📄 License

This model is fine-tuned from [stvlynn/Qwen-7B-Chat-Cantonese], which follows the Qwen license.
The dataset is derived from a public GitHub repository for academic purposes.
Please use this model in accordance with the license terms of the base model and dataset.

---

✨ Citation

If you use this model in your research or applications, please consider citing:

@misc{cantosent2025,
  author = {Geng Le},
  title = {CantoSent-Qwen-7B-OpenRice-v3: A Cantonese Sentiment Classifier},
  year = {2025},
  howpublished = {\url{https://huggingface.co/charlottegl/CantoSent-Qwen-7B-OpenRice-v3}},
  note = {Fine-tuned from Qwen for sentiment analysis on OpenRice reviews}
}

---

📬 Contact

For questions or collaborations, feel free to reach out via Hugging Face or GitHub.