Update README.md
Browse files
README.md
CHANGED
@@ -30,4 +30,96 @@ configs:
|
|
30 |
data_files:
|
31 |
- split: test
|
32 |
path: data/test-*
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
data_files:
|
31 |
- split: test
|
32 |
path: data/test-*
|
33 |
+
license: cc-by-4.0
|
34 |
+
task_categories:
|
35 |
+
- visual-question-answering
|
36 |
+
- question-answering
|
37 |
+
language:
|
38 |
+
- en
|
39 |
+
tags:
|
40 |
+
- Document Retrieval
|
41 |
+
- VisualQA
|
42 |
+
- QA
|
43 |
+
size_categories:
|
44 |
+
- 1K<n<10K
|
45 |
---
|
46 |
+
|
47 |
+
## Dataset Description
|
48 |
+
|
49 |
+
This is the test set taken from the [TAT-DQA dataset](https://nextplusplus.github.io/TAT-DQA/)is a large-scale Document VQA dataset that was constructed from publicly available real-world financial reports. It focuses on rich tabular and textual content requiring numerical reasoning. Questions and answers were manually annotated by human experts in finance.
|
50 |
+
|
51 |
+
Example of data (see viewer)
|
52 |
+
|
53 |
+
### Data Curation
|
54 |
+
Unlike other 'academic' datasets, we kept the full test set as this dataset closely represents our use case of document retrieval. There are 1,640 image-query pairs.
|
55 |
+
|
56 |
+
### Load the dataset
|
57 |
+
|
58 |
+
```python
|
59 |
+
from datasets import load_dataset
|
60 |
+
ds = load_dataset("vidore/tatdqa_test", split = 'test')
|
61 |
+
```
|
62 |
+
|
63 |
+
### Dataset Structure
|
64 |
+
|
65 |
+
Here is an example of a dataset instance structure:
|
66 |
+
|
67 |
+
```json
|
68 |
+
features:
|
69 |
+
- name: questionId
|
70 |
+
dtype: string
|
71 |
+
- name: query
|
72 |
+
dtype: string
|
73 |
+
- name: question_types
|
74 |
+
dtype: 'null'
|
75 |
+
- name: image
|
76 |
+
dtype: image
|
77 |
+
- name: docId
|
78 |
+
dtype: int64
|
79 |
+
- name: image_filename
|
80 |
+
dtype: string
|
81 |
+
- name: page
|
82 |
+
dtype: string
|
83 |
+
- name: answer
|
84 |
+
dtype: 'null'
|
85 |
+
- name: data_split
|
86 |
+
dtype: string
|
87 |
+
- name: source
|
88 |
+
dtype: string
|
89 |
+
```
|
90 |
+
|
91 |
+
## Citation Information
|
92 |
+
|
93 |
+
If you use this dataset in your research, please cite the original dataset as follows:
|
94 |
+
|
95 |
+
```latex
|
96 |
+
@inproceedings{zhu-etal-2021-tat,
|
97 |
+
title = "{TAT}-{QA}: A Question Answering Benchmark on a Hybrid of Tabular and Textual Content in Finance",
|
98 |
+
author = "Zhu, Fengbin and
|
99 |
+
Lei, Wenqiang and
|
100 |
+
Huang, Youcheng and
|
101 |
+
Wang, Chao and
|
102 |
+
Zhang, Shuo and
|
103 |
+
Lv, Jiancheng and
|
104 |
+
Feng, Fuli and
|
105 |
+
Chua, Tat-Seng",
|
106 |
+
booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)",
|
107 |
+
month = aug,
|
108 |
+
year = "2021",
|
109 |
+
address = "Online",
|
110 |
+
publisher = "Association for Computational Linguistics",
|
111 |
+
url = "https://aclanthology.org/2021.acl-long.254",
|
112 |
+
doi = "10.18653/v1/2021.acl-long.254",
|
113 |
+
pages = "3277--3287"
|
114 |
+
}
|
115 |
+
|
116 |
+
@inproceedings{zhu2022towards,
|
117 |
+
title={Towards complex document understanding by discrete reasoning},
|
118 |
+
author={Zhu, Fengbin and Lei, Wenqiang and Feng, Fuli and Wang, Chao and Zhang, Haozhou and Chua, Tat-Seng},
|
119 |
+
booktitle={Proceedings of the 30th ACM International Conference on Multimedia},
|
120 |
+
pages={4857--4866},
|
121 |
+
year={2022}
|
122 |
+
}
|
123 |
+
|
124 |
+
|
125 |
+
```
|