---
language:
- ar
license: apache-2.0
size_categories:
- 1K<n<10K
task_categories:
- text-generation
- text-classification
pretty_name: Arabic Web Educational Seed Dataset
tags:
- arabic
- education
- web-content
- quality-scoring
- educational-content
dataset_info:
  features:
  - name: text
    dtype: string
  - name: qwen2_5_72b_inst_score
    dtype: int64
  - name: prompt
    dtype: string
  - name: completion
    dtype: string
  splits:
  - name: train
    num_examples: XXXX
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---
# Arabic Web Educational Seed Dataset

## Dataset Description

This dataset contains Arabic web content that has been scored for educational value. It is designed for training language models on educational content in Arabic.

The dataset includes web extracts from various Arabic sources (Common Crawl) that have been evaluated using an additive 5-point scoring system for educational value.

## How to Load the Dataset

### Basic Loading

```python
from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("sboughorbel/arabic-web-edu-seed")

# Access the data
print(dataset['train'][0])
```

### Load Specific Split

```python
from datasets import load_dataset

# Load only the train split
dataset = load_dataset("sboughorbel/arabic-web-edu-seed", split="train")
```

### Filter by Educational Score

```python
from datasets import load_dataset

dataset = load_dataset("sboughorbel/arabic-web-edu-seed")

# Filter high-quality educational content (score >= 3)
high_quality = dataset['train'].filter(
    lambda x: x['qwen2_5_72b_inst_score'] >= 3
)

print(f"High quality examples: {len(high_quality)}")
```

### Load Only Specific Fields

```python
from datasets import load_dataset

dataset = load_dataset("sboughorbel/arabic-web-edu-seed")

# Keep only specific columns
clean_dataset = dataset['train'].select_columns([
    'text', 
    'qwen2_5_72b_inst_score',
    'url'
])
```

### Stream Large Dataset

```python
from datasets import load_dataset

# Stream the dataset without downloading everything
dataset = load_dataset("sboughorbel/arabic-web-edu-seed", streaming=True)

# Iterate through examples
for example in dataset['train']:
    print(example['text'])
    break
```

## Dataset Fields

- **text**: The Arabic text content
- **qwen2_5_72b_inst_score**: Educational quality score (0-5)
- **prompt**: The evaluation prompt used (optional)
- **completion**: The evaluation explanation (optional)

## Educational Score Scale

The `qwen2_5_72b_inst_score` ranges from 0 to 5:

- **0**: No educational value
- **1**: Basic information with non-academic content
- **2**: Some educational elements, not aligned with standards
- **3**: Appropriate for education, introduces key concepts
- **4**: Highly relevant for grade school education
- **5**: Outstanding educational value

## Citation

If you use this dataset in your research, please cite:

```bibtex
@dataset{arabic_web_edu_seed_2024,
  author = {Majd Hawasly, Tasnim Mohiuddin, Hamdy Mubarak, Sabri Boughorbel},
  title = {ArabicWeb-Edu: Educational Quality Data for Arabic LLM Training},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/sboughorbel/arabic-web-edu-seed}}
}
```

## License

[Please specify your license]

## Contact

For questions or issues, please contact through the Hugging Face dataset repository.