HiFi-KPI: A Dataset for Hierarchical KPI Extraction from Earnings Filings
The episode discusses HiFi-KPI, a dataset for hierarchical KPI extraction from earnings filings. Hosts discuss how this structured data improves AI training by providing a standardized blueprint for financial data. They cover the paper's focus on capturing metric relationships, the proposed methods for flexible extraction with user control over detail and currency units, and the long-term impact on building more robust AI systems in finance.