ibm-research/AITQARetrieval

---
annotations_creators:
- derived
language:
- eng
license: other
license_name: aitqa-license
license_link: >-
https://github.com/IBM/AITQA/blob/master/LICENSE
multilinguality: monolingual
task_categories:
- text-retrieval
task_ids:
- document-retrieval
tags:
- table-retrieval
- text
pretty_name: AIT-QA
config_names:
- default
- queries
- corpus

dataset_info:
- config_name: default
features:
- name: qid
dtype: string
- name: did
dtype: string
- name: score
dtype: int32
splits:
- name: test
num_bytes: 91137
num_examples: 1533
- config_name: queries
features:
- name: _id
dtype: string
- name: text
dtype: string
- name: answers
sequence: string
- name: type
dtype: string
- name: row_hierarchy_needed
dtype: string
- name: paraphrase_group
dtype: string
splits:
- name: test_queries
num_bytes: 52874
num_examples: 515
- config_name: corpus
features:
- name: _id
dtype: string
- name: title
dtype: string
- name: text
dtype: string
splits:
- name: corpus
num_bytes: 13506199
num_examples: 1937

configs:
- config_name: default
data_files:
- split: test
path: test_qrels.jsonl
- config_name: queries
data_files:
- split: test_queries
path: test_queries.jsonl
- config_name: corpus
data_files:
- split: corpus
path: corpus.jsonl
---

# AIT-QA Retrieval

This dataset is part of a Table + Text retrieval benchmark. Includes queries and relevance judgments across test split(s), with corpus in 1 format(s): `corpus`.

## Configs

| Config | Description | Split(s) |
|---|---|---|
| `default` | Relevance judgments (qrels): `qid`, `did`, `score` | `test` |
| `queries` | Query IDs, text, answers, type, row hierarchy flag, and paraphrase group | `test_queries` |
| `corpus` | Plain text corpus: `_id`, `title`, `text` | `corpus` |

## TableIR Benchmark Statistics

| Dataset | Structured | #Train | #Dev | #Test | #Corpus |
|---|:---:|---:|---:|---:|---:|
| OpenWikiTables | ✓ | 53.8k | 6.6k | 6.6k | 24.7k |
| NQTables | ✓ | 9.6k | 1.1k | 1k | 170k |
| FeTaQA | ✓ | 7.3k | 1k | 2k | 10.3k |
| OTT-QA (small) | ✓ | 41.5k | 2.2k | -- | 8.8k |
| MultiHierTT | ✗ | -- | 929 | -- | 9.9k |
| AIT-QA | ✗ | -- | -- | 515 | 1.9k |
| StatcanRetrieval | ✗ | -- | -- | 870 | 5.9k |
| watsonxDocsQA | ✗ | -- | -- | 30 | 1.1k |

## Citation

If you use **TableIR Eval: Table-Text IR Evaluation Collection**, please cite:

```bibtex
@misc{doshi2026tableir,
title = {TableIR Eval: Table-Text IR Evaluation Collection},
author = {Doshi, Meet and Boni, Odellia and Kumar, Vishwajeet and Sen, Jaydeep and Joshi, Sachindra},
year = {2026},
institution = {IBM Research},
howpublished = {https://huggingface.co/collections/ibm-research/table-text-ir-evaluation},
note = {Hugging Face dataset collection}
}
```

All credit goes to original authors. Please cite their work:
```bibtex
@misc{katsis2021aitqa,
title={AIT-QA: Question Answering Dataset over Complex Tables in the Airline Industry},
author={Yannis Katsis and Saneem Chemmengath and Vishwajeet Kumar and Samarth Bharadwaj and Mustafa Canim and Michael Glass and Alfio Gliozzo and Feifei Pan and Jaydeep Sen and Karthik Sankaranarayanan and Soumen Chakrabarti},
year={2021},
eprint={2106.12944},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```