Unverified50% confidenceFactExact time
CRAG（纠正式检索增强生成）会在生成前对检索结果进行质量评估和筛选，Self-RAG让模型在生成过程中自我反思引用的可靠性
1
Sources
50%
Confidence
Long-term
Relevance
9/4/2026
First Seen
Sources
Related Entities
Related Claims
Unverified开源框架RAGAS提供了RAG系统的自动化评估方案，常用指标包括检索召回率、答案忠实度和答案相关性70% similarUnverified生产级RAG评估应拆解为检索质量（Context Precision、Context Recall）、生成质量（Faithfulness、Answer Relevancy）和端到端质量三个维度67% similarUnverifiedRadO 2.0基准将置信度和人工交接能力纳入评分，部分模型会在错误诊断上给出中高置信度67% similarUnverified语义评估通常依赖NLI模型的事实一致性检测、RAG场景中的grounding评估和LLM-as-a-Judge等方法的组合65% similarUnverifiedFew-shot示例、Chain-of-Thought和RAG（检索增强生成）技术可以显著提升AI生成测试用例的业务贴合度64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/852985API
curl https://kongchang.com/api/v1/knowledge/claims/852985MCP
get_claim(id=852985)