Privacy-Preserving Fuzzy Data Matching Using Vector Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing secure intersection algorithms in privacy computing technology only match data when both parties' data are exactly the same, limiting their usage scenarios and affecting the service scope of data matching.
Innovation Solution
A data matching method using a pre-trained vector transformation model and an oblivious transfer (OT) protocol to generate intermediate vectors and distance results, enabling fuzzy matching even when data is not identical, ensuring security by keeping raw data within devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional secure intersection algorithms are used, then data security is maintained, but matching capability is limited to exact matches only
Solution Approach 1:
The patent introduces vector representations as an intermediary form between raw data and matching results. Raw data is transformed into vectors through pre-trained models, and operations are performed on these vectors rather than the original data. This intermediary representation enables fuzzy matching capabilities while the oblivious transfer protocol ensures that even the vectors do not reveal underlying data, thus resolving the contradiction between security and matching versatility.
Solution Approach 2:
The patent changes the parameter space by transforming discrete, exact-match-oriented data into continuous vector representations. This parameter transformation allows for distance-based fuzzy matching (using metrics like cosine similarity or Euclidean distance) while maintaining security through cryptographic protocols. The shift from exact string matching to vector distance measurement expands adaptability without compromising security.
2Adaptability or versatility
If fuzzy matching is implemented, then usage scenarios are broadened, but data security may be compromised
Solution Approach 1:
The oblivious transfer protocol acts as a cryptographic intermediary that enables fuzzy matching operations without exposing raw data. Parties can compute distances between their data representations and perform fuzzy matching, but the protocol ensures that neither party learns the other's underlying data. This resolves the security concern while enabling broader usage scenarios including approximate matching, similarity search, and pattern recognition.
Solution Approach 2:
The patent replaces traditional mechanical data comparison methods with cryptographic and mathematical operations. Instead of directly comparing data strings or values, the system uses vector embeddings combined with oblivious transfer protocols and distance metrics. This substitution enables fuzzy matching capabilities while maintaining security through mathematical guarantees rather than mechanical data handling.
3Measurement precision
If exact match algorithms are used, then matching precision is high, but service scope is limited
Solution Approach 1:
The patent transforms the matching parameter from binary (exact match/no match) to continuous (distance-based similarity scores). By using vector representations and distance metrics, the system can provide graded matching results indicating degrees of similarity. This parameter change maintains precision for exact matches while extending service scope to include approximate matches, partial matches, and similarity-based queries that were previously impossible.
Solution Approach 2:
The vector-based matching system provides universal functionality that encompasses both exact matching and fuzzy matching within a single framework. The same vector representation and distance computation mechanism handles both cases: when distance is zero or below a strict threshold, it indicates exact or near-exact matches; when distance is larger but below a permissive threshold, it indicates fuzzy matches. This multi-functionality expands service scope while maintaining appropriate precision through threshold selection.
Data Source
AI summary
The disclosure provides a data matching method, and the method includes inputting to-be-matched first data into a pre-trained vector transformation model, to obtain a first vector corresponding to the first data and a binary vector corresponding to the first vector; obtaining a first intermediate vector using an oblivious transfer protocol, and obtaining a first distance intermediate result based on the first intermediate vector and a first predefined algorithm; receiving a second distance intermediate result generated, based on a second vector, a first random vector, and a second predefined algorithm, and sent by a second device; determining a target distance between the first vector and the second vector based on the first vector, the first distance intermediate result, and the second distance intermediate result; and determining whether the first data and second data match based on the target distance and a predefined first distance threshold.