NovelQA

About NovelQA

News

Data Description

Each data point in our dataset is represented as a dictionary with the following keys:

[
{
    "QID": the QID which remains unchanged for tracking updates (only happen if necessary),
    "Aspect": the question classification in 'aspect', e.g., "times",
    "Complexity": the question classification in complexity, e.g., "mh",
    "Question": the input question,
    "Options": {
        "A": Option A,
        "B": Option B,
        "C": Option C (not applicable in several yes/no questions),
        "D": Option D (not application in several yes/no questions)
    },
},
...
]
           

Here is an example of a data point:

[
{
    "QID": "Q0148",
    "Aspect": "times",
    "Complex": "mh",
    "Question": "How many times has Robert written letters to his sister?",
    "Options": {
        "A": "11",
        "B": "9",
        "C": "12",
        "D": "10"
    },
},
...
]
            

Contributors

Cunxiang Wang*, Ruoxi Ning*, Boqi Pan, Tonghui Wu, Qipeng Guo, Cheng Deng, Guangsheng Bao, Qian Wang, and Yue Zhang

License

This dataset is released under the Apache-2.0 License.

Citation

@misc{wang2024novelqa,
  title={NovelQA: A Benchmark for Long-Range Novel Question Answering}, 
  author={Cunxiang Wang and Ruoxi Ning and Boqi Pan and Tonghui Wu and Qipeng Guo and Cheng Deng and Guangsheng Bao and Qian Wang and Yue Zhang},
  year={2024},
  eprint={2403.12766},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}