minorproject-research/researchy_questions_modified

Dataset Summary

This dataset is derived from corbyrosset/researchy_questions, a collection of ~100k non-factoid, multi-perspective "Researchy Questions" mined from real Bing search engine logs. After a labor-intensive filtering funnel from billions of queries, these "needles in the haystack" are questions that probably require a lot of sub-questions and research to answer adequately, and are shown to be harder than other open-domain QA datasets like Natural Questions.

Modification made: the original dataset nested the headers, subquestions, and queries fields inside a single gpt4_decomposition object per row. This dataset simply flattens that nested gpt4_decomposition column into three top-level columns — headers, subquestions, and queries — alongside the original question field. No values were changed, filtered, or regenerated; the underlying data is exactly as in the source dataset, only restructured for easier direct access.

Dataset Structure

Data Splits

Split	Examples
train	90,000
test	6,448

Data Fields

• question (string): the original user-issued search query / question.
• headers (list of list of string): grouped top-level sub-question headers from GPT-4's closed-book decomposition of the question (flattened from gpt4_decomposition.headers).
• subquestions (list of string): the flat list of sub-questions GPT-4 generated to decompose the question (flattened from gpt4_decomposition.subquestions).
• queries (list of string): search queries GPT-4 suggested for retrieving grounding documents to answer the question (flattened from gpt4_decomposition.queries).

Note: this version does not include intrinsic_scores, DocStream, decompositional_score, or nonfactoid_score from the original dataset — only question and the flattened gpt4_decomposition fields are retained.

Source Data

This dataset is a direct restructuring of corbyrosset/researchy_questions, released under the CDLA-2.0 permissive license.

Citation

If you use this dataset, please cite both the original Researchy Questions paper and this dataset.

Original dataset:

@misc{rosset2024researchy,
      title={Researchy Questions: A Dataset of Multi-Perspective, Decompositional Questions for LLM Web Agents}, 
      author={Corby Rosset and Ho-Lam Chung and Guanghui Qin and Ethan C. Chau and Zhuo Feng and Ahmed Awadallah and Jennifer Neville and Nikhil Rao},
      year={2024},
      eprint={2402.17896},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

This dataset:

@misc{[your_citation_key],
      title={[Your Dataset Title]},
      author={Aatifa Rizvi and Lakshya Varshney},
      year={2026},
      howpublished={\url{https://huggingface.co/datasets/minorproject-research/researchy_questions_modified}},
      note={Flattened version of corbyrosset/researchy\_questions}
}

License

This dataset inherits the CDLA-2.0 license of the original corbyrosset/researchy_questions dataset.

Acknowledgements

This dataset is a lightly restructured version of Researchy Questions, created by Corby Rosset and colleagues. All credit for the underlying question collection, filtering methodology, and GPT-4 decomposition annotations goes to the original authors.