基于动态红绿名单的自适应文本水印方法

An adaptive text watermarking method based on dynamic red and green lists and exemption rules solves the problems of text quality degradation and low detection rate in existing technologies, and achieves efficient text watermark generation and detection.

CN122087788BActive Publication Date: 2026-07-17BEIJING UNIV OF POSTS & TELECOMM
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
BEIJING UNIV OF POSTS & TELECOMM
Filing Date
2026-02-11
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing red and green list watermarking technologies suffer from problems such as reduced text quality due to hard red lists and low detection rates and poor robustness of soft red lists, making it difficult to strike a balance between ensuring text quality and watermark detection rate.

Method used

An adaptive method using dynamic red and green lists is adopted. Red and green lists of terms are generated through hash keys, and exemption rules are introduced to dynamically adjust the selection of candidate words during the generation process to ensure text fluency. At the same time, a high detection rate is achieved through strict exemption rules during the detection phase.

Benefits of technology

It effectively balances text quality and watermark strength, ensuring the readability of the generated text and the accuracy of detection, improving the watermark detection rate and reducing system overhead.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122087788B_ABST
    Figure CN122087788B_ABST
Patent Text Reader

Abstract

本发明公开了基于动态红绿名单的自适应文本水印方法,属于自然语言处理与人工智能安全领域,具体为:首先,用户按需求输入提示词到大语言模型,输出生成文本A;然后,获取文本每个位置的Top‑k个候选词和红绿名单。接着,逐个判断各候选词是否与绿名单有交集;如果有,从Top‑k个候选词里选出与绿名单交集中概率最高的词作为输出。否则,Top‑k个候选词全在红名单,则启动豁免规则,直接选择Top‑1的词作为等效绿名单输出;文本添加水印过程结束;最后,统计等效绿名单词元的比例进而得到统计量,验证文本是否存在水印;本发明在保障生成文本流畅性与可读性的同时,维持高水印检测率。
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • CN119337344A

  • CN121351045A