Word Edit Distance Embedding for Secure Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in securely anonymizing and classifying personal information for monitoring and compliance with regulations like GDPR, as they struggle to efficiently compare and classify words without exposing sensitive data or compromising privacy.
Innovation Solution
A neural network-based system that uses word edit distance embedding to securely anonymize words by training convolutional neural networks to approximate edit distance metrics, allowing for classification without storing original words, and employing similarity metrics for efficient searching and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional word comparison methods are used to classify personal information, then classification accuracy can be maintained, but security and privacy are compromised due to exposure of sensitive data
Solution Approach 1:
The patent creates vector representations (embeddings) of words that capture semantic meaning without exposing the original sensitive text. These vector copies allow comparison and classification operations to be performed on the encoded representations rather than the raw personal information, maintaining accuracy while protecting privacy.
Solution Approach 2:
The patent introduces vector embeddings as an intermediary between the original sensitive data and the classification process. The neural network models act as mediators that transform personal information into mathematical representations, enabling accurate classification without direct exposure of the underlying sensitive data.
2Measurement precision
If exact word matching is used for classification, then precision can be maintained, but the system becomes inefficient and cannot handle variations or typos in personal information
Solution Approach 1:
The patent transforms discrete word matching into continuous vector space comparisons. By changing the parameter space from exact string matching to geometric distance measurements in embedding space, the system can efficiently handle variations, typos, and semantic similarities while maintaining classification precision through distance-based thresholds.
3Reliability
If comprehensive data storage is maintained for accurate classification, then classification capability is preserved, but security risks and compliance costs increase
Solution Approach 1:
The patent extracts only the essential semantic features of personal information into compact vector representations, discarding the need to store the original sensitive data. This extraction approach maintains classification capability while eliminating the security risks and compliance burdens associated with storing comprehensive personal information.
Data Source
AI summary
A system for classifying words in a batch of words can include at least one memory device storing instructions for causing at least one processor to create dictionary vectors for each of a plurality of dictionary words using a neural network (NN), store each dictionary vector along with a classification indicator corresponding to the associated dictionary word, and create word vectors for each word in a batch of words for classification using the NN. The closest matching dictionary vectors are found for each word vector and the classification indicators of the closest matching dictionary vector for each word vector in the batch is reported.