Datasets:

CGU-Widelab
/

Cloze_QA_Dataset_Wikitext2

language:
  - en
tags:
  - question-answering
  - cloze
  - text
  - knowledge-based
  - QA
license: cc-by-sa-4.0
task_categories:
  - question-answering
  - text-classification

Cloze QA Dataset (WikiText-2)

Dataset Description

The Cloze QA Dataset is automatically generated from the WikiText-2 corpus. It contains fill-in-the-blank (cloze) style questions derived directly from sentences in Wikipedia articles. This dataset is particularly useful for evaluating local recall, reading comprehension, and contextual understanding.

Each document produces exactly three unique QA pairs, preserving document structure and sentence alignment while avoiding redundancy. QA pairs are stored in JSONL format, with each entry tied to a specific sentence.

Dataset Structure

Split	Documents	QA Pairs	Description
Train	5,135	15,405	Used for model training and evaluation
Validation	502	1,506	Used for hyperparameter tuning
Test	569	1,707	Used for final performance benchmarking

Example Record

{
  "doc_id": 0,
  "sent_id": 8,
  "title": "Robert Boulter",
  "question": "He appeared on a 2006 episode of the television series , ____ ,",
  "answer": "Doctors"
}

File Structure

cloze_qa_dataset/
├── train/
│   └── qa.jsonl
├── val/
│   └── qa.jsonl
└── test/
    └── qa.jsonl

Usage

This dataset is suitable for:

Training QA models (extractive or generative)
Benchmarking trivia-style QA tasks
Knowledge-based reasoning research

Source

Collected and curated by WIDELab – Web Information & Data Engineering Laboratory, Department of Computer Science and Information Engineering, Chang Gung University, Taiwan. For more information, visit CGU-Widelab.

License

Citation

@misc{CGU-Widelab/Cloze_QA_Dataset_Wikitext2,
  title={Cloze_QA_Dataset_Wikitext2},
  author={WIDELab – Web Information & Data Engineering Laboratory, Chang Gung University},
  year={2025},
  howpublished={\url{https://huggingface.co/datasets/CGU-Widelab/Cloze_QA_Dataset_Wikitext2}},
  note={Accessed: 2025-10-25}
}

@inproceedings{chaithra2025optimizingrag,
  title={Optimizing Retrieval in RAG Systems with Reinforcement Learning: A Trade-off Between Quality and Cost},
  author={Mahadevaswamy, Chaithra Lokasara and Nguyen, Khoa and Singh, Mayank and Chang, Hsien-Tsung},
  booktitle={Proceedings of the 9th International Conference on Natural Language Processing and Information Retrieval (NLPIR 2025)},
  year={2025},
  address={Fukuoka, Japan}
}