Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -1,3 +1,237 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
1 |
+

|
2 |
+
|
3 |
+
# RightNow Arabic LLM Corpus
|
4 |
+
|
5 |
+
**One of the largest high-quality Arabic text datasets for Large Language Model training and fine-tuning**
|
6 |
+
|
7 |
+
[](https://opensource.org/licenses/Apache-2.0)
|
8 |
+
[](https://en.wikipedia.org/wiki/Arabic_language)
|
9 |
+
[](#dataset-statistics)
|
10 |
+
[](#dataset-statistics)
|
11 |
+
[](https://github.com/RightNow-AI/rightnow-arabic-llm-corpus)
|
12 |
+
[](https://huggingface.co/datasets/Jr23xd23/rightnow-arabic-llm-corpus)
|
13 |
+
|
14 |
---
|
15 |
+
|
16 |
+
## Overview
|
17 |
+
|
18 |
+
This repository contains a comprehensive Arabic text dataset specifically curated for training and fine-tuning Large Language Models (LLMs). The dataset represents one of the largest collections of high-quality Arabic text available for machine learning research and commercial applications.
|
19 |
+
|
20 |
+
**Collected by [RightNow AI](https://rightnowai.co/) team** - RightNow AI is the first GPU-powered AI code editor, providing 180x more powerful AI assistance for your entire codebase.
|
21 |
+
|
22 |
+
## Dataset Statistics
|
23 |
+
|
24 |
+
| Metric | Value |
|
25 |
+
|--------|--------|
|
26 |
+
| **Total Articles** | 743,288 |
|
27 |
+
| **Total Words** | 244,153,780 |
|
28 |
+
| **Total Sentences** | 12,392,064 |
|
29 |
+
| **Unique Words** | 1,529,064 |
|
30 |
+
| **Vocabulary Richness** | 0.0063 |
|
31 |
+
| **Average Words/Article** | 328.5 |
|
32 |
+
| **Average Sentences/Article** | 16.7 |
|
33 |
+
| **Average Words/Sentence** | 19.7 |
|
34 |
+
| **High Quality Articles** | 185,351 (β₯70% quality score) |
|
35 |
+
| **Dataset Size** | 8.7GB (JSONL) |
|
36 |
+
|
37 |
+
## Key Features
|
38 |
+
|
39 |
+
- **Unprecedented Scale**: 744K articles with 244M words - one of the largest Arabic datasets available
|
40 |
+
- **Production-Ready Quality**: Advanced cleaning pipeline removes artifacts, references, and templates
|
41 |
+
- **LLM-Optimized Format**: JSONL structure designed specifically for modern language model training
|
42 |
+
- **Comprehensive Coverage**: Diverse content spanning history, science, literature, politics, and more
|
43 |
+
- **Linguistic Excellence**: UTF-8 encoded with proper Arabic text normalization and validation
|
44 |
+
- **Rich Metadata**: Complete article information including titles, URLs, and processing timestamps
|
45 |
+
|
46 |
+
## Repository Structure
|
47 |
+
|
48 |
+
```
|
49 |
+
βββ README.md # This file
|
50 |
+
βββ arabic_wikipedia_cleaned.jsonl # Main dataset (8.7GB)
|
51 |
+
βββ arabic_wikipedia_cleaned.txt # Human-readable format (8.5GB)
|
52 |
+
βββ dataset_metadata.json # Dataset metadata
|
53 |
+
βββ dataset/ # Individual cleaned files
|
54 |
+
β βββ arabic_text_*.jsonl # 11,880 individual files
|
55 |
+
β βββ ...
|
56 |
+
βββ analysis_reports/ # Comprehensive analysis
|
57 |
+
βββ dataset_analysis_*.json # Detailed JSON analysis
|
58 |
+
βββ dataset_report_*.txt # Human-readable report
|
59 |
+
βββ dataset_summary_*.csv # Summary statistics
|
60 |
+
βββ dataset_documentation_*.md # Markdown documentation
|
61 |
+
```
|
62 |
+
|
63 |
+
## Content Distribution
|
64 |
+
|
65 |
+
| Topic | Articles | Percentage |
|
66 |
+
|-------|----------|------------|
|
67 |
+
| General | 316,527 | 42.6% |
|
68 |
+
| History | 228,884 | 30.8% |
|
69 |
+
| Geography | 170,062 | 22.9% |
|
70 |
+
| Science | 118,536 | 15.9% |
|
71 |
+
| Religion | 104,378 | 14.0% |
|
72 |
+
| Politics | 87,366 | 11.8% |
|
73 |
+
| Arts | 78,915 | 10.6% |
|
74 |
+
| Literature | 76,566 | 10.3% |
|
75 |
+
| Sports | 71,171 | 9.6% |
|
76 |
+
|
77 |
+
## Quality Assessment
|
78 |
+
|
79 |
+
| Quality Level | Articles | Percentage |
|
80 |
+
|---------------|----------|------------|
|
81 |
+
| **Excellent** (β₯80%) | 130,373 | 17.5% |
|
82 |
+
| **Good** (60-80%) | 306,526 | 41.2% |
|
83 |
+
| **Fair** (40-60%) | 117,976 | 15.9% |
|
84 |
+
| **Filtered Out** (<40%) | 188,413 | 25.3% |
|
85 |
+
|
86 |
+
**Average Quality Score: 58.3%**
|
87 |
+
|
88 |
+
## Usage
|
89 |
+
|
90 |
+
### Loading the Dataset
|
91 |
+
|
92 |
+
**Available on [Hugging Face Hub](https://huggingface.co/datasets/Jr23xd23/rightnow-arabic-llm-corpus)**
|
93 |
+
|
94 |
+
**Python (Recommended)**
|
95 |
+
```python
|
96 |
+
import json
|
97 |
+
|
98 |
+
# Load the main dataset
|
99 |
+
articles = []
|
100 |
+
with open('arabic_wikipedia_cleaned.jsonl', 'r', encoding='utf-8') as f:
|
101 |
+
for line in f:
|
102 |
+
article = json.loads(line)
|
103 |
+
articles.append(article)
|
104 |
+
|
105 |
+
print(f"Loaded {len(articles)} articles")
|
106 |
+
```
|
107 |
+
|
108 |
+
**Hugging Face Datasets**
|
109 |
+
```python
|
110 |
+
from datasets import load_dataset
|
111 |
+
|
112 |
+
# Load from Hugging Face Hub
|
113 |
+
dataset = load_dataset('Jr23xd23/rightnow-arabic-llm-corpus')
|
114 |
+
|
115 |
+
# Or load from local files
|
116 |
+
dataset = load_dataset('json', data_files='arabic_wikipedia_cleaned.jsonl')
|
117 |
+
```
|
118 |
+
|
119 |
+
### Data Format
|
120 |
+
|
121 |
+
Each line in the JSONL file contains:
|
122 |
+
```json
|
123 |
+
{
|
124 |
+
"id": "unique_article_id",
|
125 |
+
"title": "Article Title",
|
126 |
+
"text": "Clean Arabic text content...",
|
127 |
+
"url": "source_url",
|
128 |
+
"hash": "content_hash",
|
129 |
+
"metadata": {
|
130 |
+
"language": "ar",
|
131 |
+
"source": "Multiple Sources",
|
132 |
+
"cleaned": true,
|
133 |
+
"processing_date": "2025-01-23T01:00:00"
|
134 |
+
}
|
135 |
+
}
|
136 |
+
```
|
137 |
+
|
138 |
+
## Use Cases
|
139 |
+
|
140 |
+
This dataset is perfect for:
|
141 |
+
|
142 |
+
- **Arabic LLM Training**: GPT, BERT, T5, LLaMA models
|
143 |
+
- **Fine-tuning**: Domain-specific Arabic models
|
144 |
+
- **Text Generation**: Content generation systems
|
145 |
+
- **NLP Research**: Arabic language processing research
|
146 |
+
- **Language Modeling**: Statistical language models
|
147 |
+
- **Transfer Learning**: Pre-trained model adaptation
|
148 |
+
|
149 |
+
## Data Processing Pipeline
|
150 |
+
|
151 |
+
Our comprehensive processing pipeline ensures high-quality data:
|
152 |
+
|
153 |
+
1. **Source Collection**: Aggregated from multiple high-quality Arabic sources
|
154 |
+
2. **Content Cleaning**: Removed artifacts, references, templates
|
155 |
+
3. **Quality Filtering**: Applied strict quality criteria (β₯70% Arabic content)
|
156 |
+
4. **Length Filtering**: Removed very short or overly long content
|
157 |
+
5. **Deduplication**: Eliminated duplicate and near-duplicate content
|
158 |
+
6. **Validation**: Comprehensive format and encoding validation
|
159 |
+
7. **Analysis**: Detailed statistical analysis and quality assessment
|
160 |
+
|
161 |
+
## Dataset Metrics
|
162 |
+
|
163 |
+
### Length Distribution
|
164 |
+
- **Article Length**: 7 - 20,757 words (median: 106)
|
165 |
+
- **Sentence Length**: 1 - 3,131 words (average: 21)
|
166 |
+
- **Word Length**: 2 - 137 characters (average: 4.9)
|
167 |
+
|
168 |
+
### Language Quality
|
169 |
+
- **Arabic Content**: β₯70% Arabic characters per article
|
170 |
+
- **Encoding**: UTF-8 with proper Arabic support
|
171 |
+
- **Text Quality**: Professional cleaning and normalization
|
172 |
+
|
173 |
+
## Technical Specifications
|
174 |
+
|
175 |
+
- **Format**: JSONL (JSON Lines)
|
176 |
+
- **Encoding**: UTF-8
|
177 |
+
- **Language**: Arabic (ar)
|
178 |
+
- **Size**: 8.7GB compressed
|
179 |
+
- **Articles**: 743,288 unique articles
|
180 |
+
- **Vocabulary**: 1.5M unique words
|
181 |
+
- **Processing Date**: January 2025
|
182 |
+
|
183 |
+
## License
|
184 |
+
|
185 |
+
This dataset is released under the **Apache License 2.0**. See the [LICENSE](LICENSE) file for the full license text.
|
186 |
+
|
187 |
+
## Contributing
|
188 |
+
|
189 |
+
We welcome contributions to improve the dataset quality and coverage:
|
190 |
+
|
191 |
+
1. **Quality Issues**: Report data quality problems
|
192 |
+
2. **Enhancement Suggestions**: Propose improvements
|
193 |
+
3. **Additional Sources**: Suggest new high-quality Arabic sources
|
194 |
+
4. **Processing Improvements**: Contribute to cleaning algorithms
|
195 |
+
|
196 |
+
## Contact
|
197 |
+
|
198 |
+
**RightNow AI** - The first GPU-powered AI code editor, providing 180x more powerful AI assistance
|
199 |
+
|
200 |
+
- Website: [https://rightnowai.co/](https://rightnowai.co/)
|
201 |
+
- Documentation: [https://docs.rightnowai.co/](https://docs.rightnowai.co/)
|
202 |
+
- Discord: [Join our community](https://discord.com/invite/sSJqgNnq6X)
|
203 |
+
- Twitter/X: [@rightnowai_co](https://twitter.com/rightnowai_co)
|
204 |
+
- Issues: [GitHub Issues](https://github.com/RightNow-AI/rightnow-arabic-llm-corpus/issues)
|
205 |
+
|
206 |
+
## Acknowledgments
|
207 |
+
|
208 |
+
This corpus was collected through the collaborative efforts of:
|
209 |
+
|
210 |
+
- **[RightNow AI](https://rightnowai.co/)** - The first GPU-powered AI code editor, providing 180x more powerful AI assistance for your entire codebase
|
211 |
+
|
212 |
+
Special thanks to:
|
213 |
+
- The Arabic language community for content creation
|
214 |
+
- Open source contributors for processing tools
|
215 |
+
- The machine learning community for feedback and validation
|
216 |
+
- RightNow AI for their dedication to advancing Arabic NLP research
|
217 |
+
|
218 |
+
## Citation
|
219 |
+
|
220 |
+
If you use this dataset in your research, please cite:
|
221 |
+
|
222 |
+
```bibtex
|
223 |
+
@dataset{rightnow_arabic_llm_corpus_2025,
|
224 |
+
title={RightNow Arabic LLM Corpus},
|
225 |
+
author={RightNow AI team},
|
226 |
+
year={2025},
|
227 |
+
publisher={GitHub},
|
228 |
+
url={https://github.com/RightNow-AI/rightnow-arabic-llm-corpus},
|
229 |
+
note={One of the largest high-quality Arabic text datasets for LLM training}
|
230 |
+
}
|
231 |
+
```
|
232 |
+
|
233 |
---
|
234 |
+
|
235 |
+
**If this dataset helps your research or project, please star this repository!**
|
236 |
+
|
237 |
+
*Built for the Arabic AI community by [RightNow AI](https://rightnowai.co/)*
|