---
license: mit
language:
- he
size_categories:
- 10K<n<100K
---
# About Dataset

The following dataset is a collection of `JSON` sets that contain most of the Hebrew Wikipedia.

## About Data collection methodology

The strategy behind the methodology for collecting the data is as follows:

- **Crawl Hebrew Wikipedia:** Begin by crawling through Hebrew Wikipedia to collect all redirect links on each page.
- **Breadth-First Search (BFS):** For each page, apply a BFS-like strategy to ensure that every link is scraped.
- **Link Collection:** Collect the links as tuples in a link file, where each tuple contains a page name and its corresponding link (<page_name>, <page_link>).
- **Data Scraping and Cleaning:** After crawling and link collection, iterate over the links and scrape all paragraphs. Apply cleaning methodologies to ensure data cleanliness. Additionally, collect the page summary for each page, if available.


### Description of the data

The dataset file is a JSONL file structured as follows:

#### JSONL File Structure

```jsonl
  {
    id: <id_num>,
    page: <page_name>,
    url: <page_url>,
    summary: <page_summary>,
    paragraphs: [<paragraph1>, <paragraph2>,...]
  }
```

| Parameter | Type     | Description                |
| :-------- | :------- | :------------------------- |
| `id` | `string` | Unique identifier for the JSON element. |
| `page` | `string` | Name of the Wikipedia page. |
| `url` | `string` | URL of the associated Wikipedia page. |
| `summary` | `string` | Uncleaned page summary, which may contain `LaTeX` for mathematical formulas or other languages besides Hebrew or English. If no summary is available, the value is `None`. |
| `paragraphs` | `string` | List of strings containing cleaned paragraphs from the page. All paragraphs only contain Hebrew and English characters. |



## Authors

- [@Fren Yan](https://github.com/YanFrenklakh)

- [@Swisa Doron](https://github.com/fkzx8000)


## License

[MIT](https://choosealicense.com/licenses/mit/)