A method and system for controllable editing of long text

CN122133622APending Publication Date: 2026-06-02BEIJING KNOWLEDGE ATLAS TECHNOLOGY CO LTD

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: BEIJING KNOWLEDGE ATLAS TECHNOLOGY CO LTD
Filing Date: 2026-03-03
Publication Date: 2026-06-02

AI Technical Summary

Technical Problem

Existing large language models lack local controllability in long text editing, struggle to balance generation quality and inference efficiency, and lack risk identification and differentiated optimization mechanisms for generated results, leading to text structure drift, semantic inconsistency, and wasted computational resources.

Method used

By splitting long texts into multiple text blocks, generating draft documents using an autoregressive model, identifying high-risk areas by combining generation confidence, text structure features, and user editing intent, and employing block-level diffusion refinement and freeze strategies, localized controllable editing and differentiated optimization are achieved.

Benefits of technology

It significantly improves the stability and controllability of long text generation, reduces inference costs and latency, and enhances generation quality and engineering deployability, making it suitable for professional writing and enterprise-level content generation.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122133622A_ABST

Patent Text Reader

Abstract

This invention belongs to the field of artificial intelligence technology and relates to a method and system for controllable editing of long text. The method includes: S1: splitting the long text into multiple text blocks; S2: generating context for each text block; S3: generating a draft document for each text block; S4: scoring the draft documents of each text block and determining whether each text block is a high-risk text block based on the scoring results; S5: for high-risk text blocks, performing multi-step denoising local rewriting and optimization of their draft documents based on the diffusion generation idea; S6: for non-high-risk text blocks, adopting a freezing strategy to ensure that their draft documents remain unchanged, and replacing the draft documents of the corresponding high-risk text blocks with the diffusion-refined documents to form a controllable edited long text. This method effectively improves the controllability and stability of long text generation without changing model parameters and training processes, while significantly reducing inference costs and system latency.

Need to check novelty before this filing date? Find Prior Art