HiFi-KPI: A Dataset for Hierarchical KPI Extraction from Earnings Filings
Researchers have introduced HiFi-KPI, a new dataset designed to improve the extraction of Key Performance Indicators (KPIs) from financial earnings reports. The dataset, comprising 1.65 million paragraphs and nearly 200,000 hierarchical labels, aims to enhance the transferability of KPI tagging across different companies. Initial evaluations show that encoder-based models achieve high accuracy in classification, while large language models demonstrate moderate success in structured extraction, with errors often related to date discrepancies. AI
IMPACT Enhances NLP capabilities for financial data analysis, potentially improving investment strategies.