Privacy-Preserving Queryable Database Using NLP Entity Abstraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing queryable databases that provide aggregated data can be vulnerable to attacks where an attacker can deduce individual information by combining queries, even with differential privacy measures, as the noise generation algorithm can be reverse-engineered, compromising privacy.

Innovation Solution

A computer-implemented method using a trained Natural Language Processing (NLP) system to assign abstract named entities to free text documents with precision and recall values between 0.75 and 1, and a ratio of 0.7 to 1.3, eliminating the need for artificial noise generation, thereby ensuring privacy through natural recognition and disambiguation errors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If differential privacy with artificial noise generation is used, then individual privacy is protected, but the noise generation algorithm can be reverse-engineered through multiple queries, compromising privacy

Engineering Contradiction:
Improveprivacy protectionVSAvoidvulnerability to attack
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts the noise generation process from the database query system itself and relocates it to the client side. The server only provides clean aggregated data, while each client independently generates noise based on their own private random seeds. This separation removes the vulnerable centralized noise generation algorithm that could be reverse-engineered, as the actual noise generation never occurs on the server where queries are processed.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces client-side random seed generation as an intermediary between the database and the privacy protection mechanism. Instead of the server generating and managing noise directly, each client acts as an independent intermediary that generates its own noise locally. This intermediary layer prevents attackers from analyzing server responses to reverse-engineer the noise generation process, as each client's noise is independently generated and never revealed to the server.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If precision and recall are set between 0.75 and 1 with ratio 0.7 to 1.3, then acceptable accuracy is maintained, but perfect precision and recall cannot be achieved

Engineering Contradiction:
Improveaccuracy of aggregated resultsVSAvoidprivacy protection
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent changes the fundamental parameter of how noise is introduced - not through adjustable precision/recall thresholds in traditional differential privacy, but through client-side random seed generation. The NLP system processes text with its inherent probabilistic nature, and the randomness is embedded in the client's local noise generation rather than being controlled by server-side parameters. This maintains accuracy while providing privacy through a different mechanism that doesn't require sacrificing precision.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4081924B1Privacy preservation in a queryable database built from unstructured texts
Publication Date: 2023.10.18 MEDSAVANA
  • EP4081924B1 patent drawingFigure 1
  • EP4081924B1 patent drawingFigure 2~3
  • EP4081924B1 patent drawingFigure 4

AI summary

A computer-implemented method of generating a queryable database (109). The method receives a corpus of free text documents (120) containing confidential data, the free text documents being related to the same domain. A trained Natural Language Processing (NLP) system (104) assigns one or more abstract named entities to each free text document in the corpus. The abstract named entities of each free text document are stored in a queryable database configured to provide aggregated information regarding the named entities. The NLP system is configured such that the abstract named entities are recognised and disambiguated with a precision between 0.75 and less than 1 and a recall between 0.75 and less than 1, and such that the ratio of precision and recall is between 0.7 and 1.3; wherein the queryable database is free from the addition of artificial noise by an artificial noise generation algorithm.