The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

MoreHopQA: More Than Multi-hop Reasoning

We propose a new multi-hop dataset, MoreHopQA, which shifts from extractive to generative answers. Our dataset is created by utilizing three existing multi-hop datasets: HotpotQA, 2Wiki-MultihopQA, and MuSiQue. Instead of relying solely on factual reasoning, we enhance the existing multi-hop questions by adding another layer of questioning.

Dataset Details

Dataset Description

Our dataset is created through a semi-automated process, resulting in a dataset with 1118 samples that have undergone human verification.
For each sample, we share our 6 evaluation cases, including the new question, the original question, all the necessary subquestions, and a composite question from the second entity to the final answer (case 3 above). We share both a version where each question was verified by a human, and a larger, solely automatically generated version ("unverified"). We recommend to primarily use the human-verified version.

  • Curated by: Aizawa Lab, National Institute of Informatics (NII), Tokyo, Japan
  • Language(s) (NLP): English
  • License: The MorehopQA dataset is licensed under CC BY 4.0

Dataset Sources

Uses

We provide our dataset to the community and hope that other researchers find it a useful tool to analyze and improve the multi-hop reasoning capabilities of their models. MoreHopQA is designed to challenge systems with complex queries requiring synthesis from multiple sources, thereby advancing the field in understanding and generating nuanced, context-rich responses. Additionally, we aim for this dataset to spur further innovation in reasoning models, helping to bridge the gap between human-like understanding and AI capabilities.

Dataset Structure

We share both a version where each question was verified by a human, and a larger, solely automatically generated version ("unverified"). We recommend to primarily use the human-verified version, which is also the default option when loading the dataset.

Each sample in the dataset contains the following fields:

  • question: Our new multi-hop question with added reasoning (case 1 above)
  • answer: The answer to the last hop (case 1, 3 and 4 above)
  • context: Relevant context information to answer the previous question (relevant for all cases except case 4)
  • previous_question: The previous 2-hop question from the original dataset (case 2 above)
  • previous_answer: The answer to the previous 2-hop question (case 2 and 5 above)
  • question_decomposition: Each question of the reasoning chain. List of entries with keys "sub_id" (position in the chain), "question", "answer", "paragraph_support_title" (relevant context paragraph). (sub_id 1 → case 6; sub_id 2 → case 5; sub_id 3 → case 4)
  • question_on_last_hop: Question for case 3 above
  • answer_type: Type of the expected answer
  • previous_answer_type: Type of the answer to the previous 2-hop question
  • no_of_hops: Number of extra hops to answer the additional reasoning question (might be more than one for more complicated tasks)
  • reasoning_type: Might contain "Symbolic", "Arithmetic", "Commonsense"; depending on which kind of reasoning is required for the additional reasoning

Dataset Creation

Source Data

Our dataset is created by utilizing three existing multi-hop datasets: HotpotQA, 2Wiki-MultihopQA, and MuSiQue

Citation

If you find this dataset helpful, please consider citing our paper

BibTeX:

@misc{schnitzler2024morehopqa,
      title={MoreHopQA: More Than Multi-hop Reasoning}, 
      author={Julian Schnitzler and Xanh Ho and Jiahao Huang and Florian Boudin and Saku Sugawara and Akiko Aizawa},
      year={2024},
      eprint={2406.13397},
      archivePrefix={arXiv}
}
Downloads last month
4