Dataset Overview

Dataset Description

Dataset Summary

HybridQA is a large‑scale QA dataset that requires reasoning over heterogeneous information. Each question is associated with a Wikipedia table and several free‑text passages linked to entities in the table. The questions are meant to aggregate both table and text information; lacking either makes the question impossible to answer.

Supported Tasks and Leaderboards

[More information to be added]

Language

The dataset is in English.

Dataset Structure

Data Instances

A typical data instance includes the following fields:

• question_id (string)
• question (string)
• table_id (string)
• answer_text (string)
• question_postag (string)
• table (dictionary):
  • url (string)
  • title (string)
  • header (list of strings)
  • data (list of dictionaries):
    • value (string)
    • urls (list of dictionaries):
      • url (string)
      • summary (string)
• section_title (string)
• section_text (string)
• uid (string)
• intro (string)

Data Splits

• Training set: 62,682 instances
• Validation set: 3,466 instances
• Test set: 3,463 instances

Dataset Creation

Rationale

[More information to be added]

Source Data

[More information to be added]

Annotation

[More information to be added]

Usage Considerations

Societal Impact

[More information to be added]

Bias Discussion

[More information to be added]

Known Limitations

[More information to be added]

Additional Information

Curators

[More information to be added]

License

The dataset is licensed under the Creative Commons Attribution 4.0 International License.

Citation

@article{chen2020hybridqa, title={HybridQA: A Dataset of Multi‑Hop Question Answering over Tabular and Textual Data}, author={Chen, Wenhu and Zha, Hanwen and Chen, Zhiyu and Xiong, Wenhan and Wang, Hong and Wang, William}, journal={Findings of EMNLP 2020}, year={2020} }

Contributions

Thanks to @patil-suraj for adding this dataset.