Add dataset card with source attribution

README.md

@@ -13,52 +13,17 @@ size_categories:
 - 1K<n<10K
 source_datasets:
 - original
-dataset_info:
-  features:
-  - name: text
-    dtype: string
-  - name: source_page
-    dtype: string
-  - name: category
-    dtype: string
-  - name: entities
-    list: string
-  - name: word_count
-    dtype: int64
-  - name: is_got_specific
-    dtype: bool
-  splits:
-  - name: train
-    num_bytes: 1953608
-    num_examples: 2820
-  - name: test
-    num_bytes: 194295
-    num_examples: 314
-  download_size: 1220795
-  dataset_size: 2147903
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: test
-    path: data/test-*
 ---
 
 # Game of Thrones Fandom Wiki Corpus
 
-A structured corpus of **3,134 prose passages** scraped from the
+A structured corpus of **3,134 prose passages** from the
 [Game of Thrones Fandom Wiki](https://gameofthrones.fandom.com/wiki)
 via its official MediaWiki API.
 
-Built as a training and evaluation resource for **LLM unlearning research**
-(see *"Who's Harry Potter? Approximate Unlearning in LLMs"*, Eldan & Russinovich 2023),
-but useful for any GoT NLP task.
-
 ## Source
 
-All text is sourced from **https://gameofthrones.fandom.com/wiki** and is
-licensed under [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/).
+All text is sourced from **https://gameofthrones.fandom.com/wiki/Wiki_of_Westeros**.
 
 ## Dataset Structure
 
@@ -87,13 +52,9 @@ licensed under [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/).
 
 ## Intended Use
 
-- Training a **reinforced model** for GoT-domain LLM unlearning
-- Evaluating whether an unlearned model retains GoT knowledge
+- Training a **reinforced model** for GoT-domain knowledge
 - GoT-domain language modeling, NER, and entity linking experiments
 
-## Scraping
+## Licence
 
-Collected with `scrape_got.py` using the MediaWiki API at
-`https://gameofthrones.fandom.com/api.php`.
-Infoboxes, tables, citations, and navigation elements were stripped.
-Paragraphs shorter than 40 words were discarded.
+[CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/).