SchemaRAG: Dynamic Large Schema Reduction for LLM-driven Structured Information Extraction
SchemaRAG:面向LLM驱动的结构化信息提取的动态大规模模式缩减
AI总结 针对大规模模式导致LLM提取结构化信息时成本高、性能下降的问题,提出SchemaRAG框架,通过检索增强动态缩减输出模式空间,在医疗和电商数据集上实现F1提升8.8%、延迟降低47%、令牌成本降低48%。
Journal ref Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track), pages 1114-1127, San Diego, California, USA, July 2026. Association for Computational Linguistics