FinGround：通过原子性声明验证检测和支撑金融幻觉
FinGround: Detecting and Grounding Financial Hallucinations via Atomic Claim Verification
AI总结：
FinGround通过三阶段验证-支撑流程，针对金融文档问答中的计算错误进行验证和修正，有效降低幻觉率，其在金融领域具有重要应用价值。
摘要：
金融AI系统必须生成基于特定监管文件的答案，但当前LLMs会编造指标、虚构引用并错误计算衍生量。这些错误具有直接监管后果，随着欧盟AI法案的高风险执行期限临近（2026年8月）。现有幻觉检测器对所有声明一视同仁，无法识别43%需要与结构化表格重新验证的计算错误。我们提出了FinGround，一种三阶段验证-支撑流程用于金融文档问答。第一阶段在文本和表格上执行金融意识混合检索。第二阶段将答案分解为原子性声明，按六种金融分类进行分类，并通过类型导向策略验证，包括公式重构。第三阶段用段落和表格单元格级别的引用重写不支持的声明。为清晰隔离验证价值与检索质量，我们提出检索等效评估作为RAG验证研究的标准方法：当所有系统接收相同检索时，FinGround仍能将幻觉率降低68%（p<0.01）。完整流程相对于GPT-4o减少78%。一个8B的简化检测器在每条声明的延迟降低18倍的情况下，保持91.4%的F1，支持0.003/查询的部署，由四周分析师试点的定性信号支持。
英文摘要：
Financial AI systems must produce answers grounded in specific regulatory filings, yet current LLMs fabricate metrics, invent citations, and miscalculate derived quantities. These errors carry direct regulatory consequences as the EU AI Act's high-risk enforcement deadline approaches (August 2026). Existing hallucination detectors treat all claims uniformly, missing 43% of computational errors that require arithmetic re-verification against structured tables. We present FinGround, a three-stage verify-then-ground pipeline for financial document QA. Stage 1 performs finance-aware hybrid retrieval over text and tables. Stage 2 decomposes answers into atomic claims classified by a six-type financial taxonomy and verified with type-routed strategies including formula reconstruction. Stage 3 rewrites unsupported claims with paragraph- and table-cell-level citations. To cleanly isolate verification value from retrieval quality, we propose retrieval-equalized evaluation as standard methodology for RAG verification research: when all systems receive identical retrieval, FinGround still reduces hallucination rates by 68% over the strongest baseline ($p < 0.01$). The full pipeline achieves a 78% reduction relative to GPT-4o. An 8B distilled detector retains 91.4% F1 at 18x lower per-claim latency, enabling $0.003/query deployment, supported by qualitative signals from a four-week analyst pilot.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
