Add docs

docs/UVW-2026.md

@@ -0,0 +1,166 @@
+# UVW 2026: Underthesea Vietnamese Wikipedia Dataset
+
+> Vietnamese Wikipedia corpus for NLP research and applications
+
+## Overview
+
+UVW 2026 (Underthesea Vietnamese Wikipedia) is a cleaned and processed dataset of Vietnamese Wikipedia articles. It is designed to support various Vietnamese NLP tasks including:
+
+- Language modeling
+- Text generation
+- Pretraining for downstream tasks
+- Text classification
+- Information retrieval
+
+## Dataset Statistics
+
+| Metric | Value |
+|--------|-------|
+| Source | Vietnamese Wikipedia |
+| Language | Vietnamese (vi) |
+| License | CC BY-SA 4.0 |
+| Year | 2026 |
+
+## Data Format
+
+### JSONL Format
+
+Each line contains a JSON object with the following fields:
+
+```json
+{
+  "id": "Việt_Nam",
+  "title": "Việt Nam",
+  "content": "Việt Nam, tên chính thức là Cộng hòa Xã hội chủ nghĩa Việt Nam...",
+  "num_chars": 45000,
+  "num_sentences": 500
+}
+```
+
+### Plaintext Format
+
+Individual text files in `plaintext/` directory, one file per article.
+
+## Data Splits
+
+| Split | Ratio |
+|-------|-------|
+| train | 80% |
+| dev | 10% |
+| test | 10% |
+
+## Usage
+
+### With Underthesea
+
+```python
+from underthesea import DataFetcher
+
+# Download and load dataset
+corpus = DataFetcher.load_corpus("UVW_2026")
+```
+
+### With HuggingFace Datasets
+
+```python
+from datasets import load_dataset
+
+# Load from HuggingFace Hub
+dataset = load_dataset("undertheseanlp/UVW-2026")
+
+# Access splits
+train_data = dataset["train"]
+val_data = dataset["validation"]
+test_data = dataset["test"]
+```
+
+### Direct Usage
+
+```python
+import json
+
+# Load JSONL
+with open("data/processed/uvw_2026.jsonl", "r", encoding="utf-8") as f:
+    for line in f:
+        article = json.loads(line)
+        print(article["title"])
+```
+
+## Building from Source
+
+### Prerequisites
+
+```bash
+pip install requests tqdm datasets
+```
+
+### Build Pipeline
+
+```bash
+# Run complete pipeline
+python scripts/build_dataset.py
+
+# Or run individual steps:
+python scripts/download_wikipedia.py    # Download Wikipedia dump
+python scripts/extract_articles.py      # Extract and clean articles
+python scripts/create_splits.py         # Create train/dev/test splits
+python scripts/prepare_huggingface.py   # Prepare for HuggingFace
+python scripts/analyze_dataset.py       # Generate statistics
+```
+
+## Processing Details
+
+### Included Articles
+
+- Main namespace articles
+- Articles with at least 100 characters of content
+
+### Excluded Content
+
+- Special pages (Wikipedia:, User:, Template:, Module:, etc.)
+- Redirect pages
+- Disambiguation pages
+- Wikipedia markup (templates, references, categories, etc.)
+- HTML tags
+- Tables
+
+### Text Cleaning
+
+1. Remove Wikipedia templates (`{{...}}`)
+2. Remove categories and files
+3. Remove references and citations
+4. Remove HTML tags
+5. Convert internal links to plain text
+6. Remove bold/italic markup
+7. Unicode normalization (NFC)
+
+## Related Datasets
+
+| Dataset | Description | Year |
+|---------|-------------|------|
+| UVB | Vietnamese Books | 2026 |
+| UVN | Vietnamese News | 2026 |
+| UTS_VLC | Vietnamese Legal Corpus | 2026 |
+| UUD | Universal Dependencies | 2026 |
+
+## Citation
+
+```bibtex
+@misc{uvw2026,
+  title={UVW 2026: Underthesea Vietnamese Wikipedia Dataset},
+  author={Underthesea NLP},
+  year={2026},
+  url={https://github.com/undertheseanlp/underthesea}
+}
+```
+
+## License
+
+This dataset is released under [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/), following the Wikipedia license.
+
+## Links
+
+- [GitHub Issue](https://github.com/undertheseanlp/underthesea/issues/896)
+- [Underthesea](https://github.com/undertheseanlp/underthesea)
+- [Vietnamese Wikipedia](https://vi.wikipedia.org)
+- [HuggingFace Dataset](https://huggingface.co/datasets/undertheseanlp/UVW-2026)

docs/underthesea_integration.md

@@ -0,0 +1,83 @@
+# Underthesea Integration
+
+This document describes how to integrate UVW 2026 into the underthesea project.
+
+## 1. Add to datasets.yaml
+
+Add the following entry to `underthesea/datasets.yaml`:
+
+```yaml
+UVW_2026:
+  cache_dir: datasets/UVW_2026
+  type: Plaintext
+  license: Open
+  year: 2026
+  filepath: ''
+  url: https://github.com/undertheseanlp/underthesea/releases/download/resources/UVW_2026.zip
+  url_filename: UVW_2026.zip
+```
+
+## 2. Create Release Package
+
+Package the processed data for release:
+
+```bash
+cd data/processed
+zip -r UVW_2026.zip plaintext/ uvw_2026.jsonl metadata.json
+```
+
+Upload `UVW_2026.zip` to GitHub releases.
+
+## 3. Usage with DataFetcher
+
+```python
+from underthesea import DataFetcher
+
+# List available datasets
+DataFetcher.list()
+
+# Download and load corpus
+corpus = DataFetcher.load_corpus("UVW_2026")
+
+# Access data
+for doc in corpus:
+    print(doc)
+```
+
+## 4. HuggingFace Hub Upload
+
+```bash
+# Login to HuggingFace
+huggingface-cli login
+
+# Create dataset repository
+huggingface-cli repo create UVW-2026 --type dataset --organization undertheseanlp
+
+# Upload parquet files
+cd data/huggingface/uvw_2026_parquet
+huggingface-cli upload undertheseanlp/UVW-2026 . --repo-type dataset
+
+# Upload README
+cd ..
+huggingface-cli upload undertheseanlp/UVW-2026 README.md --repo-type dataset
+```
+
+## 5. Integration Tests
+
+Add test to `tests/test_corpus.py`:
+
+```python
+def test_uvw_2026():
+    from underthesea import DataFetcher
+
+    corpus = DataFetcher.load_corpus("UVW_2026")
+    assert corpus is not None
+
+    # Check first document
+    docs = list(corpus)
+    assert len(docs) > 0
+```
+
+## 6. Documentation
+
+Add dataset documentation to `docs/datasets/UVW-2026.md` in the underthesea repository (see `docs/UVW-2026.md` in this project).