Hugging Face's logo

Datasets:
TheFinAI
/
FinCriticalED

Tasks:
Image-to-Text
Languages:
English
Size:
n<1K
Tags:
finance
License:
Dataset card Files
xet
Community

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.

Dataset Card for FinCriticalED Dataset

Dataset Summary

The FinCriticalED dataset contains 2 parts:

(1) raw_put: Raw HTML from financial documents retrieved from SEC EDGAR company filings and the corresponding images of the HTML chunk derived from intermediate rendered pdfs; and

(2) AnnotationWithSpan:Annotated and post-processed dataset that contains the raw image input, human expert annotated information on fincially meaningful numeric and temporal entities in the gold standard HTML format of the financial documents in (1).Each txt file represents one post-processed HTML gold standard file for one image.

This dataset is used for FinCriticalED framework on evaluating Large Language Models ability on identifying fincnanciallly critical values in complex financial documents, such as profit and revenue, important legal complaiance due dates, contract durations etc.

Supported Tasks

Task

  • Task: Image-to-Text
  • Evaluation Metrics: Financial Fact Accuracy (FFA)

Language

  • English

Dataset Structure

Data Fields

(1) raw_put

  • image : base64 encoded images of financial documents 2023-2025
  • matched_html: corresponding html that matches the corresponding image

(2) AnnotationWithSpan

Each txt file contains one post-processed HTML gold standard file for one image, the suffix id corresponds to id in raw_put

Dataset Creation

Curation Rationale

The FinCriticalED dataset was curated to support research and development on key information extraction techniques and factual reliability for unstructured documents in financial domain. By providing annotated real-world financial documents in unstructured format with ground truth, the dataset seeks to address challenges in miantaining factual consistency in complex financial OCR results.

Source Data

Initial Data Collection and Normalization

  • The source data are company filings from SEC EDGAR system.
  • Subset of company filings in 2023-2025 are downloaded in html format, and converted into correspondig images from intermediate rendered pdfs.
  • The extracted text are used for matching html chunk and correct image.

Who are the Source Language Producers?

Annotations

Annotation Process

  • The dataset was prepared by collecting, matching, spliting, and converting financial documents in English
  • Annnotators are provided with HTML chunks, rendered in Label Studio, along with its corresponding image as assistance
  • The annotators aare asked to label critical numeric and temporal entities (including dates and durations) on HTML by referrring to a carefully curated annotation guideline.

Who are the Annotators?

  • The dataset stems from publicly available financial documents.
  • The annotator team contains finance researchers and data scientists.

Personal and Sensitive Information

  • The FinCriticalED dataset does not contain any personally identifiable information (PII) and is strictly focused on English-language open-source financial documents. No personal or sensitive information is present in the dataset.

Considerations for Using the Data

Social Impact of Dataset

This dataset enables AI models to extract structured information from scanned financial documents, promoting transparency and accessibility. By aligning page-level images with HTML data accurately labeled with financially critical numerical and temporal entities, it supports the development of more factual reliable and accurate financial OCR work in the finance domain.

Discussion of Biases

  • The source data is limited to financial documents, it may underrepresent other domains, potentially limiting model generalizability.
  • The data may over-represent English financial contexts.
  • The data emphasizes on numerical and temporal data, may underrepresent other critical financial informations, such as company names, person names, etc.

Other Known Limitations

  • While the dataset covers wide range of financial documents, it may lack sufficient variety in layout styles (e.g., handwritten notes, non-standard financial forms, embedded charts), which could limit a model’s ability to generalize to less structured or unconventional financial documents.

Additional Information

Dataset Curators

  • Yueru He
  • The FinAI Team

Licensing Information

  • License: Apache License 2.0

Citation Information

If you use this dataset, please cite:

@misc{
}
Downloads last month
16