面向文档知识库的多粒度结构化检索增强生成方法及装置
By parsing and maintaining the structural information between document fragments in the retrieval enhancement generation system, adaptive recall and reordering are achieved, solving the problems of difficulty in adaptive information granularity and inaccurate matching in existing systems, and realizing more accurate information services.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INST OF SOFTWARE - CHINESE ACAD OF SCI
- Filing Date
- 2024-04-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing retrieval enhancement generation systems disrupt the multiple information granularities and structural relationships of documents during index building, leading to difficulties in information granularity adaptation and inaccurate matching of effective information, making it difficult to meet users' information needs at different granularities.
Before building the index, the system parses and maintains the structural information between document fragments. Through hierarchical combination relationships, referential relationships, and native reference relationships, it adaptively recalls information fragments of different granularities and uses structured information to reorder them, thereby improving the accuracy of information matching.
It enables the effective utilization of multi-granularity information and structural relationships within and outside the document, adaptively recalls information of appropriate granularity, improves the accuracy of information matching, avoids the risk of error propagation, and provides more accurate information services.
Smart Images
Figure CN118585615B_ABST
Abstract
Citation Information
Patent Citations
Intelligent text information processing system
CN115455935A
Questions and answers generation
US20110125734A1