GB/Z 253-2026Artificial intelligence - General technical requirements of retrieval-augmented generation (English PDF)

Foreword

This document is a standard or guiding technical document.

This document complies with the provisions of GB/T 1.1-2020 "Standardization Work Guidelines Part 1.Structure and Drafting Rules of Standardization Documents".

Drafting.

Please note that some content in this document may involve patents. The issuing organization of this document assumes no responsibility for identifying patents.

This document was proposed and is under the jurisdiction of the National Information Technology Standardization Technical Committee (SAC/TC28).

1 Scope

This document establishes a general technical framework for search enhancement generation and specifies the functional requirements for search enhancement generation, knowledge base, and other related components.

This document is intended to guide the design and development of retrieval enhancement generation systems.

2 Normative references

This document has no normative references.

3 Terms and Definitions

The following terms and definitions apply to this document.

3.1

retrieval-augmented generation

Methods for enhancing generative artificial intelligence models on specific tasks by retrieving relevant information from pre-built knowledge bases or data sources.

An artificial intelligence system built on retrieval-enhanced generation technology.

4 Abbreviations

The following abbreviations apply to this document.

MP3.MPEG-1 Audio Layer 3, Version 1 of the Moving Picture Experts Group standard.

MP4.MPEG-4 Part 14, Version 4, Standard from the Moving Picture Experts Group.

RAG. Retrieval-Augmented Generation

The general technical framework of RAG is shown in Figure 1.The framework mainly includes.

a) Enhanced Retrieval Generation. Through processes such as query understanding, knowledge retrieval, semantic ranking, and answer generation, it generates responses relevant to the query question.

Answer;

b) Knowledge Base. Accessing data from different sources and heterogeneous structures, performing knowledge acquisition and segmentation, extracting entity information, and generating corresponding questions.

We construct a knowledge base related to query questions using keywords such as titles and abstracts, so that it can be indexed more efficiently through vector indexing, full-text indexing, and other methods.

Leading to.

END: Draft Version (GBZ253-2026)