Datasets:

cierpa
/

esg-qa-dataset

Modalities:

Formats:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

ESG-QA: Benchmark Dataset for Multimodal Question Answering of Japanese ESG Documents

Dataset

This dataset consists of the following two subsets:

`documents`

Metadata of source PDF documents.

document - Document ID.
sha256 - SHA256 of the PDF file.
url - Original URL of the PDF file.

This dataset does not include the raw PDF files. If you wish to obtain the PDFs from our database, please submit an application via the request form.

`annotations`

Annotations to the above PDF documents.

document - Document ID.
question - Question.
answer_type - Answer types: YES_NO, NUMBER, FREE_TEXT, EXTRACTION
answers - List of answers.
references - List of the following objects indicating references for the answer.
- page - Page number. (0-indexed)
- x0 - Left position of a bounding box.
- y0 - Top position of a bounding box.
- x1 - Right position of a bounding box.
- y1 - Bottom position of a bounding box.
expert_annotation - Additional information about the answer.

References

佐多亮明, 赤部晃一, 神田峻介, 小田悠介. ESG-QA: 日本語ESG文書を対象としたマルチモーダル質問応答ベンチマークの構築. 言語処理学会第32回年次大会 (NLP2026). 宇都宮. 2026年3月. (in Japanese) (PDF)

Downloads last month: 7