AgentKGV: Agentic LLM-RAG Framework with Two-Stage Training for the Fact Verification of Knowledge Graphs

AgentKGV：用于知识图谱事实验证的代理式 LLM-RAG 两阶段训练框架

Knowledge graphs (KGs) are often automatically constructed from large-scale corpora, but they inevitably contain factual errors due to noisy sources and extraction failures, and verifying them reliably at industrial scale remains a critical challenge. 知识图谱（KG）通常是从大规模语料库中自动构建的，但由于源数据噪声和提取失败，它们不可避免地包含事实错误。在工业规模上可靠地验证这些知识图谱仍然是一项严峻的挑战。

To address this, we propose AgentKGV, the Agentic LLM-RAG framework for KG fact Verification, that integrates dynamic routing and iterative query rewriting, which handles surface-form mismatch in document-level retrieval. 为了解决这一问题，我们提出了 AgentKGV，这是一个用于知识图谱事实验证的代理式 LLM-RAG 框架。该框架集成了动态路由和迭代查询重写功能，能够处理文档级检索中出现的表层形式不匹配问题。

To make this framework more accurate and cost-efficient for industrial deployment, we further introduce a two-stage training strategy: turn-level distillation-based SFT that transfers reasoning ability from a large teacher model into a small model for stable query rewriting and reasoning, and trajectory-level GRPO that optimizes the search policy to reduce unnecessary retrieval at scale. 为了使该框架在工业部署中更加准确且具有成本效益，我们进一步引入了两阶段训练策略：一是基于轮次级蒸馏的监督微调（SFT），将大型教师模型的推理能力迁移到小型模型中，以实现稳定的查询重写和推理；二是轨迹级 GRPO（分组相对策略优化），通过优化搜索策略来减少大规模场景下不必要的检索。

On the long-tail-predicate split of the open-domain T-REx benchmark, our framework improves macro-F1 over single-turn RAG by 5.5 %p, and two-stage training does it further by 9.4 %p. GRPO also cuts the average number of search calls from 3.24 to 1.63 without lowering accuracy. 在开放域 T-REx 基准测试的长尾谓词划分上，我们的框架将 macro-F1 指标较单轮 RAG 提升了 5.5 个百分点，而两阶段训练进一步将其提升了 9.4 个百分点。此外，GRPO 在不降低准确率的情况下，将平均搜索调用次数从 3.24 次减少到了 1.63 次。