CoolFace
Datasetpublic

didi0di/finance-legal-mrc-with-images

๐Ÿงพ finance-legal-mrc-with-images (tableqa/test) Multimodal-ready table image dataset designed for TIG (Table Information Generation) inputTIG ์ž…๋ ฅ ์ „์šฉ ํ…Œ์ด๋ธ” ์ด๋ฏธ์ง€ ๋ฐ์ดํ„ฐ์…‹ (VLM ํ™œ์šฉ ๊ฐ€๋Šฅ) ๐Ÿ“ฆ Dataset Overview | ๋ฐ์ดํ„ฐ์…‹ ๊ฐœ์š” Feature Description (EN) ์„ค๋ช… (KR) ๐Ÿงฉ Split tableqa / test tableqa / test ์Šคํ”Œ๋ฆฟ ๐Ÿ“„ Total Rows 1,197 (unique tables only) ์ด 1,197๊ฑด (์ค‘๋ณต ์ œ๊ฑฐ๋œ ๊ณ ์œ  ํ…Œ์ด๋ธ” ๊ธฐ์ค€) ๐Ÿ–ผ๏ธ Image Format PNG (rendered from raw HTML tables) ์›๋ณธ HTML ํ…Œ์ด๋ธ”์„ PNG ์ด๋ฏธ์ง€๋กœ ๋ Œ๋”๋ง ๐Ÿ”— Source Derived fromโ€ฆ See the full description on the dataset page: https://huggingface.co/datasets/didi0di/finance-legal-mrc-with-images.

sourceHugging Facecc-by-sa-4.0updated 1y agoView on Hugging Face
0likes15downloads
Dataset Card

๐Ÿงพ finance-legal-mrc-with-images (tableqa/test)

Multimodal-ready table image dataset designed for TIG (Table Information Generation) input TIG ์ž…๋ ฅ ์ „์šฉ ํ…Œ์ด๋ธ” ์ด๋ฏธ์ง€ ๋ฐ์ดํ„ฐ์…‹ (VLM ํ™œ์šฉ ๊ฐ€๋Šฅ)


๐Ÿ“ฆ Dataset Overview | ๋ฐ์ดํ„ฐ์…‹ ๊ฐœ์š”

FeatureDescription (EN)์„ค๋ช… (KR)
๐Ÿงฉ Splittableqa / testtableqa / test ์Šคํ”Œ๋ฆฟ
๐Ÿ“„ Total Rows1,197 (unique tables only)์ด 1,197๊ฑด (์ค‘๋ณต ์ œ๊ฑฐ๋œ ๊ณ ์œ  ํ…Œ์ด๋ธ” ๊ธฐ์ค€)
๐Ÿ–ผ๏ธ Image FormatPNG (rendered from raw HTML tables)์›๋ณธ HTML ํ…Œ์ด๋ธ”์„ PNG ์ด๋ฏธ์ง€๋กœ ๋ Œ๋”๋ง
๐Ÿ”— SourceDerived from `shchoice/finance-legal-mrc`๊ธฐ์กด MRC ๋ฐ์ดํ„ฐ์…‹์—์„œ ํ™•์žฅ๋จ
โš ๏ธ Note: This dataset does not include TIG outputs (e.g., title, description, summary, key_entities). Use it as input for VLM-based TIG generation. โš ๏ธ ์ฃผ์˜: ๋ณธ ๋ฐ์ดํ„ฐ์…‹์€ TIG ๊ฒฐ๊ณผ๋ฌผ(title, description ๋“ฑ)์„ ํฌํ•จํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋น„์ „-์–ธ์–ด ๋ชจ๋ธ ๊ธฐ๋ฐ˜ TIG ์ƒ์„ฑ์„ ์œ„ํ•œ ์ž…๋ ฅ ์ „์šฉ ๋ฐ์ดํ„ฐ์…‹์ž…๋‹ˆ๋‹ค.

๐Ÿง  Use Cases | ํ™œ์šฉ ์˜ˆ์‹œ

  • โ€”๐ŸŽฏ TIG Input for VLM ์ด๋ฏธ์ง€ ๊ธฐ๋ฐ˜ ํ…Œ์ด๋ธ” ์ •๋ณด๋ฅผ ์ž…๋ ฅ์œผ๋กœ ํ™œ์šฉํ•ด GPT-4o ๋“ฑ์˜ ๋ชจ๋ธ์—์„œ TIG ์ˆ˜ํ–‰
  • โ€”๐Ÿ“Š Table Structure Understanding ๋ชจ๋ธ์ด ํ…Œ์ด๋ธ” ๊ตฌ์กฐ, ํŒจํ„ด, ๋ถ„ํฌ๋ฅผ ์‹œ๊ฐ์ ์œผ๋กœ ํ•ด์„ํ•˜๋„๋ก ํ•™์Šต
  • โ€”๐Ÿ” Multimodal RAG or Retrieval Tasks ์ด๋ฏธ์ง€ ๊ธฐ๋ฐ˜ ์งˆ์˜์‘๋‹ต ๋˜๋Š” ๊ฒ€์ƒ‰(RAG) ํ”„๋ ˆ์ž„์›Œํฌ ํ…Œ์ŠคํŠธ

๐Ÿ–ผ๏ธ Sample | ์ƒ˜ํ”Œ ์˜ˆ์‹œ

unique_idtable_titletable_image
uniquetableid_0[ํ‘œ 1] ์„ธ๋Œ€๋ณ„ ์ด ์ธ๊ตฌ์ˆ˜ ๋ณ€ํ™”img
uniquetableid_1[ํ‘œ 2] ๊ณ ๋ น์šด์ „์ž ๊ตํ†ต์‚ฌ๊ณ  ์‚ฌ๋ง๋ฅ  ๋น„๊ตimg

๐Ÿ“ Data Fields | ํฌํ•จ๋œ ํ•„๋“œ

  • โ€”unique_id: ์œ ๋‹ˆํฌํ•œ ํ…Œ์ด๋ธ” ID (unique_table_id_0 ๋“ฑ)
  • โ€”origin_id: ์›๋ณธ finance-legal-mrc ๋ฐ์ดํ„ฐ์…‹์—์„œ์˜ ์ธ๋ฑ์Šค
  • โ€”document_title: ํ…Œ์ด๋ธ”์ด ํฌํ•จ๋œ ๋ฌธ์„œ ์ œ๋ชฉ
  • โ€”context: ํ…Œ์ด๋ธ” ์ฃผ๋ณ€ ๋ฌธ๋งฅ ์ •๋ณด
  • โ€”table_title: ํ‘œ ์ œ๋ชฉ
  • โ€”table_html: ์›๋ณธ HTML ํ…Œ์ด๋ธ”
  • โ€”table_image: ์‹œ๊ฐํ™”๋œ ํ…Œ์ด๋ธ” ์ด๋ฏธ์ง€ (Hugging Face viewer ์ง€์›)

โœจ How It Was Created | ์ƒ์„ฑ ๊ณผ์ •

  1. 1.finance-legal-mrc/tableqa/test์—์„œ ์ค‘๋ณต๋˜์ง€ ์•Š๋Š” ํ…Œ์ด๋ธ” 1,197๊ฐœ ์ถ”์ถœ
  2. 2.HTML ํ…Œ์ด๋ธ”์„ ์ด๋ฏธ์ง€(PNG)๋กœ ๋ Œ๋”๋ง
  3. 3.๋ฉ”ํƒ€๋ฐ์ดํ„ฐ(JSON + ์ด๋ฏธ์ง€ ๊ฒฝ๋กœ) ๊ตฌ์„ฑ ํ›„ datasets.Image ํƒ€์ž…์œผ๋กœ ์ฒ˜๋ฆฌ
  4. 4.DatasetDict๋ฅผ ๊ตฌ์„ฑํ•ด tableqa/test ์Šคํ”Œ๋ฆฟ์œผ๋กœ ์—…๋กœ๋“œ

๐Ÿ“Œ Citation

bibtex
@misc{didi0di_finance_legal_mrc_with_images,
  author = {didi0di},
  title = {Finance-Legal-MRC with Images},
  year = {2025},
  howpublished = {\url{https://huggingface.co/datasets/didi0di/finance-legal-mrc-with-images}},
  note = {TIG input dataset based on shchoice/finance-legal-mrc, extended with table images}
}

๐Ÿ›  Maintainer

  • โ€”โœ๏ธ didi0di (Hugging Face Profile)
  • โ€”๐Ÿ“ซ Questions or feedback? Feel free to open a discussion or contact me on Hugging Face!