README.md

---
license: cc-by-sa-4.0
pretty_name: Wikipedia "Did You Know" Archive
language:
  - en
  - fr
  - de
multilinguality:
  - multilingual
task_categories:
  - text-generation
  - question-answering
annotations_creators:
  - crowdsourced
language_creators:
  - crowdsourced
source_datasets:
  - original
size_categories:
  - 100K<n<1M
tags:
  - wikipedia
  - did-you-know
  - trivia
  - encyclopedic
configs:
  - config_name: en
    data_files:
      - split: train
        path: en/dyk-en.parquet
  - config_name: fr
    data_files:
      - split: train
        path: fr/dyk-fr.parquet
  - config_name: de
    data_files:
      - split: train
        path: de/dyk-de.parquet
---

# Wikipedia "Did You Know" Archive

Structured data curated from the "Did you know" (DYK) snippets Wikipedia
communities feature on the Main Page — the short, sourced "... that *X*?"
facts. In DYK parlance each such fact is a **hook**; this dataset holds one
record per hook occurrence, enriched with the articles the hook links to.

| Config | Language | Rows | DYK section |
| --- | --- | --- | --- |
| `en` | English | 117,342 | *Did you know ...* |
| `fr` | French | 12,011 | *Le saviez-vous ?* |
| `de` | German | 14,354 | *Schon gewusst?* |

The schema is identical across languages (language-specific absences are
`null`), so the configs can be mixed or compared directly.

## Usage

```python
from datasets import load_dataset

en = load_dataset("wikimedia/wiki-dyk", "en", split="train")
print(en[0]["content_body"])
```

## Data fields

One record per hook occurrence:

| Field | Type | Description |
| --- | --- | --- |
| `language` | string | Wiki language code (`en`, `fr`, `de`). |
| `archive_page` | string | Source archive page the hook was read from. |
| `day` | date | Date the hook appeared on the Main Page. `null` where the archive does not record it (French before 2017). |
| `set_time` | timestamp (UTC) | Set-removal timestamp where the archive records one (English from 2008); `null` elsewhere. |
| `hook_wikitext` | string | Raw wikitext of the hook, for re-rendering or re-analysis. |
| `content_body` | string | Reader HTML rendered locally from the wikitext (bold, italic, links). |
| `linked_pages` | list<struct> | Articles the hook links to, enriched (see below). |

Each entry in `linked_pages`:

| Field | Type | Description |
| --- | --- | --- |
| `page_id` | int64 | Page id of the canonical (redirect-resolved) target; `null` if the article was later deleted. |
| `rev_id` | int64 | Revision current on the DYK date — historically accurate and reproducible. |
| `page_title` | string | Canonical target title (redirects resolved), not the link text. |
| `short_description` | string | Short description (local on enwiki, Wikidata-backed elsewhere); `null` if none. |
| `page_image` | string | The article's free page image filename; `null` if none. |
| `wikidata_qid` | string | The article's Wikidata item (e.g. `Q42`). |
| `article_topics` | list<struct{qid:int64, title:string}> | Topic categories from the topictrends service; `qid` is the topic's Wikidata item (stable cross-language key), `title` its localized name. `null` when unavailable. |
| `prominent` | bool | The featured article(s) of the hook — the bold link(s) on wikis with a bolding convention. |

### Example record

```json
{
  "language": "en",
  "archive_page": "Wikipedia:Did you know archive/2026/July",
  "day": "2026-07-08",
  "set_time": "2026-07-08T00:00:00Z",
  "hook_wikitext": "* ... that '''[[Iron Bridge, Lucknow|one of the earliest iron bridges in India]]''' ''(pictured)'' was ...?",
  "content_body": "... that <b><a href=\"./Iron_Bridge,_Lucknow\">one of the earliest iron bridges in India</a></b> <i>(pictured)</i> was ...?",
  "linked_pages": [
    {
      "page_id": 12345,
      "rev_id": 67890,
      "page_title": "Iron Bridge, Lucknow",
      "short_description": "Bridge in Uttar Pradesh, India",
      "page_image": "Iron_bridge_Lucknow.jpg",
      "wikidata_qid": "Q130206787",
      "article_topics": [{ "qid": 9079095, "title": "Bridges_in_India" }],
      "prominent": true
    }
  ]
}
```

## How it was built

DYK archive pages are fetched as raw wikitext and parsed per-language;
`content_body` is rendered locally from the parse tree (no HTML API). Linked
pages are enriched via each wiki's action API: canonical title and page id
(redirects resolved), the revision current on the DYK date, short description,
free page image, Wikidata item, plus topic categories from the topictrends
service. A hook appearing across multiple days or sets yields one record per
occurrence.

Source code: https://github.com/santhoshtr/dyk-archive/

## License and attribution

This dataset is derived from Wikipedia article and archive text, which is
released under [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/)
(also available under the GFDL). The dataset is distributed under the same
terms.

Attribution is to the Wikipedia contributors of each source wiki
([en](https://en.wikipedia.org), [fr](https://fr.wikipedia.org),
[de](https://de.wikipedia.org)). Per CC BY-SA, reuse must credit Wikipedia and
share adaptations under a compatible license. `page_title`, `rev_id`, and
`archive_page` let any fact be traced back to its exact source revision.

## Limitations

- Coverage follows each wiki's archive completeness; months with no archive
  page are simply absent (normal gaps, not errors).
- `content_body` is a faithful local rendering, but unusual templates in old
  eras may render as raw wikitext in rare cases.
- `day` and `set_time` are `null` where the archive format never recorded them.