Multi-Party Text Watermarking Using Asymmetric Key Partitioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing generative models lack effective methods to embed watermarks for multi-party collaboration, leading to potential unauthorized use and theft of generated content, which complicates ownership verification and traceability.
Innovation Solution
A method involving asymmetric key pairs from multiple users is used to partition a glossary in a generative pre-training model, generating text content with embedded watermarks that can be verified by each user without revealing the other party's keys, ensuring secure and confidential collaboration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If watermarks are embedded in generated content to provide uniqueness and traceability, then ownership verification and content protection are improved, but the system complexity and difficulty of implementation increase due to multi-party key management requirements
Solution Approach 1:
The patent segments the watermarking system into independent key pairs for each user. Each user generates their own key pair and embeds only their public key into the generated content, rather than requiring a centralized watermarking system. This segmentation reduces system complexity while maintaining ownership verification capabilities through distributed key management.
Solution Approach 2:
The patent introduces an intermediary mechanism where public keys serve as the mediation layer between users and the watermarking system. Instead of directly embedding sensitive private keys or requiring complex multi-party cryptographic protocols, the system uses public keys as intermediaries that can be freely embedded in content while still enabling verification through their corresponding private keys.
2Measurement precision
If multiple users collaborate on content generation with individual watermarks, then traceability and ownership identification are improved, but the ease of operation deteriorates due to the need to manage and coordinate multiple key pairs
Solution Approach 1:
The patent enables self-service operation where each user independently generates their own key pair, embeds their public key into the generated content, and verifies watermarks using their private key. Users do not need to coordinate with other users or manage their keys, making the system easy to operate while maintaining precise traceability of each user's contribution.
3Reliability
If public keys are embedded in generated content for watermarking, then ownership identification is improved, but security risks increase if private keys are exposed during the process
Solution Approach 1:
The patent extracts and embeds only the public key component into the generated content, completely separating it from the private key. The private key never leaves the user's device and is never transmitted or stored externally. This extraction approach enables ownership identification through embedded watermarks while eliminating security risks associated with private key exposure.
Data Source
AI summary
A method in one embodiment includes acquiring a first public key and a first private key of a first user and a second public key and a second private key of a second user. The method further includes determining a first set from a glossary based on the first public key and the first private key, wherein the glossary is a set of a group of predefined tokens in a generative pre-training model. The method further includes determining a second set from the glossary based on the second public key and the second private key. The method further includes determining a set containing a plurality of tokens based on the first set and the second set. The method further includes generating text content by the generative pre-training model based on the set. According to embodiments of the present disclosure, a unique watermark can be formed in the generated text content.