OpenDFM/AirQA-Real

NeuSym-RAG: Hybrid Neural Symbolic Retrieval with Multiview Structuring for PDF Question Answering

This repository contains the dataset, database and vectorstore employed in the experiment part of our paper NeuSym-RAG: Hybrid Neural Symbolic Retrieval with Multiview Structuring for PDF Question Answering.

Dataset Statistics

We choose and experiment on these three datasets:

Dataset	PDF Count	Test Data Size	Task Type	Evaluation Type	Original Link
AirQA-Real	6795	553	single, multiple, retrieval	subjective, objective	this work
M3SciQA	1789	452	multiple	subjective	https://github.com/yale-nlp/M3SciQA
SciDQA	576	2937	single, multiple	subjective	https://github.com/yale-nlp/SciDQA

Folder Structure of Dataset Directory

For each dataset, we provide:

• a zip file containing relevant materials of this dataset. Its structure is as follows:

  airqa/
  ├── metadata/
  |   |── a0008a3c-743d-5589-bea2-0f4aad710e50.json
  |   └── ... # more metadata dicts
  ├── papers/
  |   |── acl2023/
  |   |   |── 001ab93b-7665-5d56-a28e-eac95d2a9d7e.pdf
  |   |   └── ... # more .pdf published in ACL 2023
  |   └── ... # other sub-folders of paper collections
  |── processed_data/
  |   |── a0008a3c-743d-5589-bea2-0f4aad710e50.json # cached data for PDF parsing
  |   └── ... # more cached data for PDFs
  |── test_data_*.jsonl
  └── uuids.json
  

  For M3SciQA dataset, there will be another sub-folder images containing the anchor images of the questions.
• a duckdb file, corresponding to the database we use in the main experiments.
• a db file, corresponding to the vectorstore we use in the main experiments.

For more information, please refer to the documents in our repository.

Citation

If you find this dataset useful, please cite our work:

@misc{cao2025neusymraghybridneuralsymbolic,
      title={NeuSym-RAG: Hybrid Neural Symbolic Retrieval with Multiview Structuring for PDF Question Answering}, 
      author={Ruisheng Cao and Hanchong Zhang and Tiancheng Huang and Zhangyi Kang and Yuxin Zhang and Liangtai Sun and Hanqi Li and Yuxun Miao and Shuai Fan and Lu Chen and Kai Yu},
      year={2025},
      eprint={2505.19754},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.19754}, 
}