Privacy-Preserving Fuzzy String Matching with Vector Distance

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing fuzzy string matching techniques incur significant computational overhead when implemented under multi-party computation, making them infeasible for practical use due to high branching, and there is a need for privacy-preserving methods that can efficiently handle spelling variations in text strings.

Innovation Solution

A cryptographic protocol using a string-to-vector model, such as word2vec, to determine vector representations of input strings outside of the multi-party computation, followed by computing distances between these representations using efficient scalar products, thereby reducing communication rounds and computational overhead.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If existing fuzzy string matching techniques are implemented under multi-party computation, then privacy preservation is achieved, but computational overhead becomes significantly high making practical use infeasible

Engineering Contradiction:
Improveprivacy preservationVSAvoidcomputational efficiency
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent applies preliminary action by pre-computing and sharing Bloom filter structures before the actual fuzzy string matching operation. The Bloom filter is constructed from the reference string and shared in advance, allowing the remote party to perform efficient filtering without needing to perform complex computations during the matching phase. This pre-preparation significantly reduces the computational overhead during the actual matching operation while maintaining privacy preservation.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent extracts the essential functionality of fuzzy string matching into a simplified Bloom filter-based approach. Instead of performing complete fuzzy string matching algorithms under MPC, the invention extracts only the critical filtering function into a compact Bloom filter structure that can be efficiently evaluated remotely. This extraction allows the system to maintain privacy while dramatically reducing computational complexity.

Inventive Principle:
Principle #2Taking out (Extraction)

2Loss of information

If multi-party computation is used for fuzzy string matching, then data privacy is protected, but communication rounds and computational overhead increase

Engineering Contradiction:
Improvedata privacyVSAvoidcommunication rounds
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The Bloom filter structure is constructed and shared in advance before the actual matching operation. This preliminary preparation of the filtering mechanism allows the remote party to perform efficient evaluation without requiring multiple communication rounds during the matching phase. The advance sharing of the Bloom filter reduces interactive communication requirements while maintaining privacy protection.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses a simplified copy of the reference string information encoded in the Bloom filter structure. Instead of sharing the actual reference string or performing complex matching algorithms, the system creates a compact representation (Bloom filter) that captures the essential filtering capability. This copying approach maintains privacy by not exposing the original data while enabling efficient remote evaluation.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4533725B1Privacy-preserving fuzzy string matching
Publication Date: 2026.03.04 ROSEMAN GRP BV
  • EP4533725B1 patent drawingFigure 1~2
  • EP4533725B1 patent drawingFigure 3
  • EP4533725B1 patent drawingFigure 4

AI summary

The invention relates to a cryptographic system (010) for performing a fuzzy string matching in a privacy-preserving way as a cryptographic secure multi-party computation between multiple cryptographic devices. Input devices apply a string-to-vector model to input strings to obtain vector representations, and provides the vector representations as inputs to the multi-party computation. A cryptographic device obtains the vector representations as inputs to the multi-party computation and computes under the multi-party computation distances between the vector representations. The distance is used as an indicator of a fuzzy string matching between the input strings.