FinEvolveKG：可控可归因的金融知识图谱提示演化
原标题：FinEvolveKG: Controlled and Attributable Prompt Evolution for Financial Knowledge Graph Construction
AI 摘要
论文提出 FinEvolveKG，一种面向金融知识图谱构建的提示词演化框架，将可复用的通用块与 persona 策略块分离，并机械地将反射式提示优化器 GEPA 限制在指定块内，使每次编辑可归因、可独立审查。在基于 SEC 10-K 的 ticker 不相交数据划分上，完整两阶段块约束流程（v4）相比冻结提示基线提升 19.4–30.9%，相比无约束全提示流程（v5）提升 2.2–6.4%，并在所有 OOD 偏移和九组配对种子比较中优于 v5。SME 审计支持将生成三元组用作 ground truth，盲审中 v4 在 70.1% 的判断中优于 v5。
正文节选
FinEvolveKG: Controlled and Attributable Prompt Evolution for Financial Knowledge Graph Construction Abstract. Financial institutions construct knowledge graphs (KGs) from shared SEC 10-Ks to serve risk analysis, portfolio management, and investment research, whose analytical needs differ. Extraction must adapt to evolving needs without fragmenting shared grounding, typing, and temporal rules. Hand-authored persona prompts scale poorly, and persona-specific fine-tuning adds cost. Weight-free pro