Semantic Document Categorization for Sensitive Data Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in accurately detecting sensitive data buried in unstructured files and assigning appropriate classification labels, leading to false positives, user errors, and inefficient data security measures.
Innovation Solution
An electronic device employs semantic categorization using Machine Learning techniques to determine the privacy persona and classification label of documents, enabling precise detection of sensitive information and automated classification without relying on patterns or user input.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If regular expression patterns and context words are used to detect sensitive data, then detection coverage is improved, but false positives increase and computational cost increases
Solution Approach 1:
The patent replaces the mechanical pattern-matching system (regular expressions and context word analysis) with a machine learning-based semantic understanding system. The ML model learns to recognize sensitive data through training on labeled examples, capturing complex semantic relationships without relying on rigid pattern templates. This substitution enables the system to distinguish true sensitive data from false positives by understanding context at a semantic level rather than a syntactic level.
Solution Approach 2:
The patent changes the fundamental parameters of detection from fixed pattern templates and static context word lists to dynamic, learned parameters from machine learning models. The system adjusts its detection criteria based on learned patterns from training data, allowing it to adapt to different data types and contexts. This parameter transformation enables more accurate detection while reducing false positives, as the model learns optimal detection thresholds and features during training.
2Measurement precision
If more specific patterns are matched against every document, then detection precision is improved, but computational cost increases
Solution Approach 1:
The patent performs preliminary action by training the machine learning model in advance on a large corpus of labeled data. During this offline training phase, the model learns to identify patterns and features associated with sensitive data. Once trained, the model can perform rapid inference on new documents without requiring intensive computational resources during the actual detection process. This preliminary learning action transfers the computational burden from the detection phase to the training phase.
Solution Approach 2:
The patent creates a simplified representation or copy of the complex detection logic through the trained machine learning model. Instead of executing complex pattern-matching algorithms against every document, the system uses the trained model's learned parameters and features to make rapid detection decisions. The model essentially copies the essential detection knowledge into a compact form that can be applied efficiently at scale.
3Measurement precision
If user-driven classification label assignment is implemented, then classification accuracy is improved, but user workload and error rates increase
Solution Approach 1:
The patent implements self-service by enabling the system to automatically assign classification labels without requiring user intervention. The machine learning model, trained on labeled examples, learns to autonomously determine appropriate classification labels for documents based on their content and characteristics. This self-service capability eliminates the need for users to manually review and assign labels, significantly reducing workload while maintaining or improving classification accuracy through consistent, error-free automated decisions.
Solution Approach 2:
The patent replaces the manual user-driven classification process with an automated machine learning-based system. Instead of relying on users to review and assign labels, the ML model automatically performs classification based on learned patterns from training data. This substitution maintains high accuracy by leveraging the model's learned understanding of classification criteria while eliminating human errors and reducing operational burden.
4Measurement precision
If comprehensive data search is performed to identify all sensitive data, then detection completeness is improved, but processing time increases
Solution Approach 1:
The patent changes the detection approach from exhaustive pattern matching to efficient machine learning-based classification. The ML model, trained to recognize various types of sensitive data, can process documents more quickly by learning to focus on relevant features and patterns. The model transforms the detection problem into a classification task that can be solved more efficiently, maintaining high detection completeness while reducing processing time through optimized inference algorithms.
Data Source
AI summary
Embodiments herein disclose a method for managing at least one document by an electronic device 100. The method includes receiving, by the electronic device, at least one document in an electronic form, where the document includes a plurality of content. Further, the method includes determining at least one semantic category associated with the document and a metadata associated with the document. Further, the method includes determining at least one among privacy persona and classification label of the document based on the semantic category associated with the document and the metadata associated with the document. Further, the method includes determining access to the document based on at least one among the privacy persona and the classification label of the document.