FinRED

Description

FinRED is an open‐source relation extraction dataset specifically curated for the financial domain, addressing the lack of domain‐focused RE resources. It contains annotated triples drawn from financial news and earnings call transcripts, aligned via distant supervision to Wikidata and manually validated for evaluation.

Dataset Creators

  • Soumya Sharma, IIT Kharagpur

  • Tapas Nayak, IIT Kharagpur

  • Arusarka Bose*, IIT Kharagpur

  • Ajay Kumar Meena*, IIT Kharagpur

  • Koustuv Dasgupta, Goldman Sachs

  • Niloy Ganguly, IIT Kharagpur & LU Hannover

  • Pawan Goyal, IIT Kharagpur

(* indicates corresponding authors)

Task Description

Relation extraction models must identify and classify semantic relationships between entity pairs in financial text, mapping them to a predefined set of relation types relevant to finance (e.g., “acquired_by”, “reported_earnings”, “appointed_ceo”).

Key characteristics: - Source texts: Financial news articles and earnings call transcripts - Annotation:

  • Training and development data aligned by distant supervision from Wikidata triplets

  • Manually annotated test set for robust evaluation

  • Relation types: A comprehensive set covering finance‐specific relations (e.g., mergers & acquisitions, executive appointments, financial performance metrics)

  • Size: Tens of thousands of auto‐labeled training instances; several hundred manually validated test instances

  • Format: JSON lines, with each record containing “head_entity”, “relation”, “tail_entity”, and sentence context

  • Licensing & Availability: Publicly released under CC BY‐NC 4.0 on GitHub and Hugging Face

Example dataset (https://huggingface.co/datasets/FinGPT/fingpt-finred-re):

Sample FinRED relation annotations

Evaluation Metrics

  1. Precision, Recall, F1 (micro and macro) over relation types

  2. Accuracy of relation classification

References

@misc{sharma2025finred,
  title        = {FinRED: A Dataset for Relation Extraction in Financial Domain},
  author       = {Sharma, Soumya and Nayak, Tapas and Bose, Arusarka and Meena, Ajay Kumar and Dasgupta, Koustuv and Ganguly, Niloy and Goyal, Pawan},
  howpublished = {\url{https://github.com/soummyaah/FinRED/}},
  year         = {2025},
  note         = {CC BY-NC 4.0}
}