Context-Aware Document Sanitization via ML Classification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing document sanitization techniques, such as find-and-replace methods, fail to accurately identify and obscure sensitive information in documents, especially non-textual content like images and audio, and struggle with context-dependent information, leading to incomplete or inaccurate redaction and potential information loss.

Innovation Solution

A context-based document content classification and alteration method that uses a content classification model trained with machine learning to identify and classify content segments within documents, determining whether they should be altered or left intact based on semantic meaning and context, and provides user interface-driven modifications to ensure accurate and context-aware redaction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional find-and-replace methods are used for document sanitization, then the process is simple and fast, but the accuracy of identifying sensitive information is low and context-dependent information cannot be properly handled

Engineering Contradiction:

Improveaccuracy of sensitive information identificationVSAvoidcomplexity of sanitization system

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary classification model between the raw document and the redaction process. This model analyzes content segments and their contexts to determine whether they should be redacted, serving as a mediator that resolves the contradiction between simple find-and-replace methods and accurate context-aware sanitization.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the mechanical find-and-replace system with an intelligent classification-based system. Instead of using rigid keyword matching, the system uses machine learning models to classify content segments based on their semantic meaning and context, thereby improving accuracy while managing complexity through automation.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If context-based classification with machine learning is implemented, then the accuracy of sensitive information identification is improved, but the processing time and computational resources increase

Engineering Contradiction:

Improveaccuracy of context-dependent information redactionVSAvoidprocessing time for document sanitization

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent divides the document into discrete content segments and processes each segment independently through the classification model. This segmentation allows the system to apply context-based analysis only where needed, reducing overall processing time while maintaining high accuracy for sensitive information identification.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies full context-based classification only to content segments that are likely to contain sensitive information, while using simpler methods for obviously non-sensitive segments. This partial application of the complex classification process reduces processing time while maintaining accuracy for critical content.

Inventive Principle:

Principle #16Partial or excessive action

3Reliability

If comprehensive redaction of all identified sensitive information is performed, then information security is improved, but the risk of information loss and document usability degradation increases

Engineering Contradiction:

Improveinformation security through redactionVSAvoidinformation loss and document usability

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The patent applies different redaction treatments to different content segments based on their classification results and contextual importance. Instead of uniformly redacting all identified sensitive information, the system selectively redacts only those segments that are truly sensitive while preserving contextually appropriate information, thereby maintaining document usability.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent incorporates user feedback mechanisms where users can review and adjust redaction decisions. The system learns from user corrections and refines its classification model, ensuring that redaction actions are both secure and preserve necessary document usability over time.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS11455464B2Document content classification and alteration

Publication Date: 2022.09.27 ACCENTURE GLOBAL SOLUTIONS LTD

AI summary

A device may receive a document including text, images, and one or more embedded documents. The device may parse the document to identify a content segment in the document. The device may determine a context for the content segment, wherein the context includes at least one of an entity identified by the content segment, a semantic meaning of the content segment, or an object identified by the content segment. The device may classify the content segment using a content classification model and based on the context for the content segment. The device may selectively alter the content segment based at least in part on a set of alteration rules, to generate a modified document. The device may provide the modified document based on selectively altering the content segment.