The authors present FinKG-News, a framework that automatically constructs company-centric financial knowledge graphs by extracting news events as anchors linked to companies. This grounded evidence integrates events, news, and company data to support an in-context learning architecture for generating credit risk reports across three core financial dimensions.
- The system explicitly models event-market relations through factual, environment-aware knowledge graphs.
- It utilizes an in-context learning architecture to generate credit risk reports based on the constructed graph.
- Automated evaluations show improvements in quality by 19%-34% while reducing hallucinations compared to baselines.
- Human evaluations indicate that automated hallucination detection and quality assessment remain unreliable, necessitating expert judgment.
The approach demonstrates consistent outperformance over baselines, though the authors note that human expertise is still required for reliable quality assessment.