Here is the dataset for supporting the project RAG_fairness, **Paper Title**  
[**Does RAG Introduce Unfairness in LLMs? Evaluating Fairness in Retrieval-Augmented Generation Systems**](https://aclanthology.org/2025.coling-main.669/)

[Official Gihub Repo](https://github.com/elviswxy/RAG_fairness)

The `RAG_fairness_corpus.tar.gz` contains the raw wiki corpus used in this project.

* `wiki_dump.jsonl`: the raw wiki dumps (14G~, 21,015,324 docs), used for building the index (bm25 or e5) for BBQ evaluation, you can get more information from official [FLASHRAG REPO](https://github.com/RUC-NLPIR/FlashRAG/blob/main/docs/original_docs/process-wiki.md) for more deatails about how to create the `wiki_dump.jsonl` file.
  * Format: '{"id": "", "contents": ""}'
* `trec_fair_2022_train_id_title_only_first_100w_clean_corpus.jsonl`: our special version for `trec-fair/2022/train` (4G~, 6,300,043 docs), used for `building the index` (bm25 or e5). You can follow our code in github for more details how to build, which based on [this implementation](https://github.com/elviswxy/RAG_fairness/blob/main/preprocess_trec_wiki.py).
  * Format: '{"id": "", "title": "", "text": ""}'
* `trec_fair_2022_train_id_title_only_first_100w_clean_corpus_formatted.jsonl`: our special version for `trec-fair/2022/train` (4G~, 6,300,043 docs), used for FlashRAG retrival format. You can follow our code in github for more details how to build, which based on [this implementation](https://github.com/elviswxy/RAG_fairness/blob/main/preprocess_trec_wiki.py).
  * Format: '{"id": "", "contents": ""}'
  
---
license: apache-2.0
---