Secure Outsourced Dataset Annotation via Encrypted Token Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for secure annotation of sensitive datasets in outsourced environments fail to meet requirements of keeping reference annotations private, preventing re-encryption, enabling unattended server operation, minimizing data leakage, and maintaining efficiency comparable to clear-text operations, particularly in scenarios involving personal identifiable information.
Innovation Solution
A method involving encryption of datasets and references, permutation of fields, and addition of dummy fields to enhance unlinkability, allowing secure annotation and reannotation without decrypting data, using symmetric deterministic encryption and random permutations to protect sensitive data from untrusted servers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the dataset is encrypted to protect privacy, then data security is improved, but the server cannot perform annotation operations
Solution Approach 1:
The patent introduces encrypted token fields and encrypted comparison fields as intermediaries that enable the server to perform annotation operations on encrypted data. The server compares encrypted tokens from the dataset with encrypted comparison fields from the reference, allowing annotation without decryption. This mediator approach resolves the contradiction by enabling computation on encrypted data.
Solution Approach 2:
The patent transforms the data representation by converting clear-text data into encrypted form while maintaining the ability to perform comparisons. By changing the parameter of data representation from clear-text to encrypted form with special handling of token fields, the system achieves both security and operational capability.
2Reliability
If the reference is re-encrypted for every query to maintain security, then data security is improved, but the processing time increases significantly
Solution Approach 1:
The patent applies preliminary encryption to the reference data before outsourcing, creating encrypted comparison fields that can be reused for multiple annotation queries. The reference is encrypted once with deterministic encryption, allowing the same encrypted reference to be used across multiple queries without re-encryption, thus resolving the time consumption issue while maintaining security.
Solution Approach 2:
The patent creates encrypted copies of the reference data that can be reused across multiple queries. Instead of re-encrypting the reference for each query, the system uses deterministic encryption to generate consistent encrypted representations that can be safely reused, eliminating redundant encryption operations.
3Productivity
If the server performs annotation on unencrypted data for efficiency, then processing speed is improved, but data leakage occurs
Solution Approach 1:
The patent uses encrypted token fields and encrypted comparison fields as intermediaries that enable the server to perform annotation operations directly on encrypted data without decryption. The server compares encrypted values and performs operations in the encrypted domain, maintaining both security and efficiency by avoiding the need to decrypt data during processing.
4Reliability
If dummy fields are added to enhance unlinkability, then privacy protection is improved, but the data structure complexity increases
Solution Approach 1:
The patent introduces dummy fields as intermediaries that enhance unlinkability by preventing the server from determining whether annotation results correspond to actual data records or dummy entries. This adds a layer of indirection that protects privacy while maintaining a manageable data structure through systematic handling of real and dummy fields.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention discloses a method and a system for outsourced annotation wherein the system comprises: a client component; and a server component wherein the client component comprises a client processor to classify contents of the dataset into token fields and zero or more fields associated to the token fields and wherein encryption module is to reorganize the dataset and encrypt the dataset, the server component being to match the encrypted dataset token fields with token fields of a reference, to define for each record with a matching token field the information fields that are to be annotated and annotating by the annotation module the corresponding associated information fields.