README.md

---
license: mit
task_categories:
- question-answering
- text-generation
language:
- vi
tags:
- vietnamese
- t5
- nlp
- question-generation
pretty_name: HVU Question Generation Dataset
size_categories:
- 10K<n<100K
---

# HVU_GQ

**HHVU_GQ** is a project dedicated to sharing datasets and tools for **Question Generation Processing (NLP)**, developed and maintained by the research team at **Hung Vuong University (HVU), Phu Tho, Vietnam**.  
This project is supported by **Hung Vuong University, Phu Tho, Vietnam**, with the aim of advancing research and applications in low-resource language processing, particularly for the Vietnamese language.

---

## 📚 Overview

This repository enables you to:
1. Fine-tune the [VietAI/vit5-base](https://huggingface.co/datasets/DANGDOCAO/GeneratingQuestions) model on your own QA dataset.
2. Generate multiple, diverse questions given a user-provided text passage (context).

---

## 📁 Dataset Format

Your dataset must follow the **SQuAD v2.0** JSON structure:

```json
{
  "version": "v2.0",
  "data": [
    {
      "title": "Đồ uống nào của Việt Nam từng lọt top ngon nhất thế giới?",
      "paragraphs": [
        {
          "qas": [
            {
              "id": "q1_1",
              "question": "Đồ uống nào của Việt Nam từng lọt top ngon nhất thế giới?",
              "answers": [
                {
                  "text": "Theo bản đánh giá tháng 2/2023 của Taste Atlas...",
                  "answer_start": "Theo bản đánh giá tháng 2/2023 của Taste Atlas..."
                }
              ],
              "is_impossible": false
            }
          ]
        }
      ]
    }
  ]
}
```

**Required fields:**
- `title` → used as context
- `question` → target question
- `answers[0].text` → seed answer for training
- `is_impossible` → filter for valid QAs

**File name:** `30ktrain.json` (UTF-8)

---

## 📁 Datasets

This repository provides datasets for **training** and **evaluating** Vietnamese question generation models.

### 🔹 `DataTotalQCAtriples30k/`
- **`30ktrain.json`** → 30,000 QCA triples for training.

### 🔹 `Datatest1k/`
- **`testorgin1k.json`** → 1,000 examples for manual & automatic evaluation.

### 🔹 `Datatrain29k/`
- **`29kcorpustag.json`** → 29,000 preprocessed QCA triples.

> All files are UTF-8 encoded and ready for direct use in NLP pipelines.

---

## 📊 Evaluation Results

We performed **manual evaluation on 500 samples** and **automatic evaluation on 1,000 samples**.

| Evaluation Type  | Precision | Recall | F1-Score |
|------------------|-----------|--------|----------|
| Automatic (1000) | 0.85      | 0.83   | 0.84     |
| Manual (500)     | 0.88      | 0.86   | 0.87     |

---

## 🔧 Installation

Create a virtual environment and install dependencies:

### Windows (PowerShell)
```powershell
python -m venv .venv
.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
pip install torch transformers datasets scikit-learn
```

### Linux / macOS
```bash
python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install torch transformers datasets scikit-learn
```

---

## 🚀 Training (Fine-tuning)

Make sure `30ktrain.json` is in the same folder as `fine_tune_qg.py`.

Run:
```bash
python fine_tune_qg.py
```

**What happens:**
- Loads `30ktrain.json`
- Splits into 80% train / 20% validation
- Tokenizes using `T5Tokenizer`
- Fine-tunes `VietAI/vit5-base` for 3 epochs
- Saves model + tokenizer to `t5-viet-qg-finetuned/`

**Key training parameters** (edit in `fine_tune_qg.py` if needed):
- `per_device_train_batch_size`: 1
- `learning_rate`: 2e-4
- `num_train_epochs`: 3
- `max_input_length`: 512
- `max_target_length`: 64

---

## 💡 Generating Questions

Once training is done, use `generate_question.py` to generate new questions.

Ensure:
- `MODEL_DIR` → `t5-viet-qg-finetuned/`
- `DATA_PATH` → `30ktrain.json`

Run:
```bash
python generate_question.py
```

Steps:
1. Enter a context passage (Vietnamese)
2. Enter number of questions (default 20, max 200)
3. Script will:
   - Find best match in dataset by title similarity
   - Use matched answer + your context
   - Generate multiple unique questions with top-k & top-p sampling
4. Output lists generated questions.

---

## ⚙️ Generation Settings

In `generate_question.py`, tweak:
- `top_k` (default 60)
- `top_p` (default 0.95)
- `temperature` (default 0.9)
- `no_repeat_ngram_size` (default 3)
- `repetition_penalty` (default 1.12)

---

## 📂 Project Structure

```
.
├── fine_tune_qg.py
├── generate_question.py
├── 30ktrain.json
├── t5-viet-qg-finetuned/
├── README.md
└── LICENSE
```

---

## 🔍 Example Usage

**Training**
```bash
python fine_tune_qg.py
```

**Generating**
```bash
python generate_question.py
```

Example:
```
Nhập đoạn văn bản:
 Cà phê sữa đá là đồ uống nổi tiếng ở Việt Nam.

Nhập vào số lượng câu hỏi bạn cần: 5

✅ Các câu hỏi mới được sinh ra:
1. Loại cà phê nào nổi tiếng ở Việt Nam?
2. Tại sao cà phê sữa đá được yêu thích?
3. Cà phê sữa đá gồm những nguyên liệu gì?
4. Nguồn gốc của cà phê sữa đá là từ đâu?
5. Cà phê sữa đá Việt Nam được pha chế như thế nào?
```

---

## 🤝 Contribution

You’re welcome to:
- Open issues
- Submit pull requests
- Suggest new datasets

---

## 📄 License

Licensed under the MIT License – see the `LICENSE` file for details.

---

## 📬 Contact

- **Ha Nguyen-Tien** (Corresponding author)  
  Email: nguyentienha@hvu.edu.vn  

- **Phuc Le-Hong**  
  Email: Lehongphuc20021408@gmail.com

- **DANG DO CAO**  
  Email: docaodang532001@gmail.com

---

*This repository is part of an effort to advance Vietnamese NLP by making question generation more accessible for researchers and developers.*