一种基于多层索引的知识检索方法

By constructing a multi-layered index structure and combining vector similarity analysis of knowledge classification tags, document summaries, and text blocks, the problems of fuzzy knowledge base classification, excessively broad retrieval scope, and contextual breaks in existing technologies are solved, achieving efficient and accurate knowledge retrieval.

CN120633843BActive Publication Date: 2026-07-17INSPUR GENERSOFT CO LTD

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
INSPUR GENERSOFT CO LTD
Filing Date
2025-05-30
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

The lack of an effective classification system in existing knowledge bases means that RAG technology needs to traverse multiple unrelated knowledge bases during retrieval, which increases retrieval complexity and time cost. Furthermore, traditional methods cannot guarantee the accuracy and contextual integrity of retrieval results.

Method used

A knowledge retrieval method based on multi-level indexing is adopted. By constructing a three-level index structure of knowledge classification tags, document summaries and text blocks, semantic analysis and vector similarity calculation are performed using a large language model. The query statement and tags are dynamically matched to narrow the search scope and quickly locate highly relevant documents and text blocks.

Benefits of technology

It significantly improves the accuracy and efficiency of knowledge retrieval, reduces redundant calculations, and ensures the contextual coherence of retrieval results and the quality of answer generation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120633843B_ABST
    Figure CN120633843B_ABST
Patent Text Reader

Abstract

本发明公开了一种基于多层索引的知识检索方法,包括:根据知识检索语句,结合知识分类标签,得到对应分类的知识库;根据知识检索语句,结合对应分类的知识库内的文档摘要,通过向量相似性分析,得到多个检索目标文档;根据知识检索语句,结合多个检索目标文档的文本块,通过向量相似性分析,得到多个相似文本块,以作为知识检索输出。本发明构建了包含知识分类标签、文档摘要、文本块的三层索引结构,通过大语言模型动态匹配分类标签缩小检索范围,结合文档摘要向量相似性分析快速定位高相关性文档,并基于文本块层级向量匹配保留上下文逻辑关系。解决了传统技术的分类模糊、文档召回低效及上下文断裂问题,实现宏观到微观的精准检索。
Need to check novelty before this filing date? Find Prior Art