Method for implementing semantic chain retrieval enhancement generation process for long document knowledge question answering

By using the Semantic Chain RAG method and knowledge graph technology, we optimized long document knowledge question answering, solved the problems of semantic mismatch and context loss, and improved the accuracy and efficiency of long document knowledge question answering.

CN121029966BActive Publication Date: 2026-06-09THE THIRD RES INST OF MIN OF PUBLIC SECURITY
View PDF 5 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
THE THIRD RES INST OF MIN OF PUBLIC SECURITY
Filing Date
2025-08-13
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing technologies suffer from semantic mismatch, context loss, and limited model context capacity in long document knowledge question answering, which affect the accuracy and efficiency of the answers.

Method used

We employ the semantic chain RAG method, combining fine-grained document segmentation, chained storage, information compression, and knowledge graph technologies. By constructing a document vector library, an information compression library, and a knowledge graph library, we perform fine-grained semantic encoding, vector similarity calculation, and entity linking to optimize the generation process of long document knowledge question answering.

Benefits of technology

It effectively solves the problems of semantic mismatch and context loss in long document retrieval, improves retrieval efficiency and semantic integrity, and enhances the ability to handle complex semantic relationships and the robustness of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121029966B_ABST
    Figure CN121029966B_ABST
Patent Text Reader

Abstract

The application relates to a method for implementing semantic chain retrieval enhancement generation processing for long document knowledge question answering, mainly including knowledge base construction and user knowledge question answering. The user knowledge question answering is based on the knowledge base construction, and the knowledge base construction involves long document preprocessing, semantic chain storage, information compression, knowledge extraction and storage. When the user performs knowledge question answering, fine-grained semantic coding is first performed into a retrieval vector, and then relevant document vectors and knowledge graphs are matched from the knowledge base through vector similarity calculation. If the document length exceeds the context scale of the model, static information compression after document information is indexed through information compression, and then it is evaluated whether dynamic compression is performed, and the document segment related to the task is returned through dynamic compression. Otherwise, the relevant original document is directly returned as the background knowledge of the generator. The model generator generates an answer based on the relevant document information / information after information compression, knowledge graph information, the user question and the model prompt.
Need to check novelty before this filing date? Find Prior Art