AI RESEARCH
FinReflectKG -- HalluBench: GraphRAG Hallucination Benchmark for Financial Question Answering Systems
arXiv CS.CL
•
ArXi:2603.20252v1 Announce Type: new As organizations increasingly integrate AI-powered question-answering systems into financial information systems for compliance, risk assessment, and decision, ensuring the factual accuracy of AI-generated outputs becomes a critical engineering challenge. Current Knowledge Graph (KG)-augmented QA systems lack systematic mechanisms to detect hallucinations - factually incorrect outputs that undermine reliability and user trust. We