README.md

---
dataset_info:
  features:
  - name: correct_text
    dtype: string
  - name: error_text
    dtype: string
  - name: __index_level_0__
    dtype: int64
  splits:
  - name: train
    num_bytes: 895314409
    num_examples: 3175684
  - name: test
    num_bytes: 111815675
    num_examples: 396961
  - name: validation
    num_bytes: 112054245
    num_examples: 396961
  download_size: 742034799
  dataset_size: 1119184329
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
  - split: validation
    path: data/validation-*
license: mit
task_categories:
- fill-mask
- text-generation
- token-classification
language:
- vi
---

## Dataset Card for Vietnamese Text Correction Dataset

## Dataset Description

This dataset contains Vietnamese text pairs for training and evaluating text correction models. Each example consists of an erroneous text and its corrected version, making it ideal for:

- **Grammar correction**
- **Spelling correction** 
- **Text normalization**
- **Language model fine-tuning**

### Dataset Summary

- **Language**: Vietnamese (vi)
- **Format**: Text correction pairs
- **Size**: ~4.0M examples across train/validation/test splits
- **Domain**: General Vietnamese text from various sources
- **Collection Method**: Automated collection and human verification

## Dataset Structure

### Data Instances

A typical data point looks like this:

```json
{
  "correct_text": "Đây là một câu tiếng Việt chuẩn xác.",
  "error_text": "Đây là một câu tieengs Việt chuẩn xác.",
  "__index_level_0__": 42
}
```

### Data Fields

- **correct_text**: The corrected Vietnamese text (string)
- **error_text**: The original text with errors (string) 

### Data Splits

| Split | Examples | Size |
|-------|----------|------|
| **train** | 3,175,684 | 895 MB |
| **test** | 396,961 | 112 MB |
| **validation** | 396,961 | 112 MB |

## Dataset Creation

### Source Data

The dataset was created by collecting Vietnamese text from multiple sources including:
- Web crawling of Vietnamese websites
- Social media posts
- News articles
- User-generated content

### Processing Steps

1. **Text Collection**: Gathered raw Vietnamese text from diverse sources
2. **Error Injection**: Applied various error types (spelling, grammar, typos)
3. **Human Annotation**: Native speakers corrected the erroneous texts
4. **Quality Filtering**: Removed low-quality or nonsensical examples
5. **Deduplication**: Ensured no duplicate text pairs

## Usage

### Loading the Dataset

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("8Opt/vn-text-correction-0001")

# Access different splits
train_data = dataset['train']
test_data = dataset['test']
val_data = dataset['validation']

# Example usage
example = train_data[0]
print(f"Original: {example['error_text']}")
print(f"Corrected: {example['correct_text']}")
```

### For Training

```python
# Prepare for training a correction model
def preprocess_function(examples):
    inputs = [f"Correct this text: {err}" for err in examples['error_text']]
    targets = examples['correct_text']
    return {'input_text': inputs, 'target_text': targets}

# Apply preprocessing
tokenized_datasets = dataset.map(preprocess_function, batched=True)
```

## Evaluation

### Metrics

Common evaluation metrics for this dataset include:
- **BLEU Score**: Measures n-gram overlap with reference corrections
- **ROUGE Score**: Evaluates summary-quality corrections
- **Character/Word Error Rate**: Traditional correction metrics
- **Human Evaluation**: Manual assessment of correction quality

## Limitations and Bias

- **Domain Coverage**: Dataset may not cover all Vietnamese domains equally
- **Error Types**: Focuses on common errors; rare error types may be underrepresented  
- **Regional Variations**: Northern/Central/Southern Vietnamese differences may exist
- **Contemporary Usage**: May not capture very recent slang or terminology

## Ethical Considerations

- All text data is publicly available Vietnamese content
- No personally identifiable information included
- Dataset intended for research and educational purposes only


## Contributing

We welcome contributions! If you find issues or want to improve the dataset:

1. Open an issue on the https://huggingface.co/datasets/8Opt/vn-text-correction-0001.
2. Submit a pull request with improvements
3. Report any data quality issues or annotation errors

## Contact

For questions or feedback about this dataset, please contact: [minh.leduc.0210@gmail.com]