Dual-Language-Model PII Scrubbing for Accuracy and Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional language models (LMs) lack accuracy and efficiency in identifying and sanitizing personally identifiable information (PII), requiring high computational resources and manual intervention, which leads to delays and limited throughput.
Innovation Solution
A dual-language model approach where a first LM is trained with ground truth labels to enhance accuracy, and a second, less resource-intensive LM is configured to perform PII scrubbing, leveraging context awareness and reduced computational demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional language models are used for PII identification and sanitization, then the system can perform basic text processing, but the accuracy in identifying PII is insufficient leading to underinclusive or overinclusive sanitization
Solution Approach 1:
The system segments the PII identification task into multiple specialized models: a primary PII detection model, a secondary verification model, and category-specific models. This segmentation allows each model to focus on specific aspects of PII identification, improving overall accuracy and reducing false positives/negatives without requiring a single overly complex model.
Solution Approach 2:
The patent introduces an intermediary verification layer where a second language model validates the PII identification results from the first model. This intermediary step acts as a mediator to confirm or correct initial detections, thereby improving reliability and reducing errors in sanitization without directly modifying the core detection mechanism.
2Productivity
If conventional language models are used for PII sanitization, then the system can process text data, but the computational resource demands are high causing unacceptable delays and limited throughput
Solution Approach 1:
The system divides the computational workload across multiple specialized models with different complexity levels. The primary model handles initial detection, while secondary verification only processes flagged items. This segmentation reduces the computational burden on any single model and enables parallel processing, improving throughput without proportionally increasing resource consumption.
Solution Approach 2:
The patent applies different levels of model complexity to different parts of the processing pipeline. Simpler models handle routine detections, while more complex verification is applied only when needed. This local quality approach ensures high accuracy where necessary while minimizing computational resources for routine operations, thereby improving overall productivity.
3Extent of automation
If conventional language models are used for PII sanitization, then the system can perform basic sanitization tasks, but manual intervention is required leading to delays and increased operational complexity
Solution Approach 1:
The system performs preliminary automated verification using the second language model before requiring manual review. This preliminary action filters out most cases that can be confidently resolved by the automated system, reducing the volume of work requiring manual intervention and thereby decreasing overall processing time while maintaining high automation levels.
Solution Approach 2:
The patent implements a feedback loop where manual review outcomes are used to retrain and improve the language models. This continuous feedback mechanism progressively reduces the need for manual intervention over time as the models become more accurate, thereby increasing automation extent while reducing the time loss associated with manual operations.
4Measurement precision
If a highly accurate first language model is trained with ground truth labels, then the PII detection accuracy improves, but the computational resource requirements increase
Solution Approach 1:
The system segments the model architecture into a primary detection model and a secondary verification model. The first model is trained for high accuracy on specific PII types, while the second model provides verification. This segmentation allows the system to achieve high overall accuracy without requiring a single extremely complex model, thereby managing device complexity while maintaining precision.
Solution Approach 2:
The patent applies the more complex first language model only when necessary for difficult cases, while simpler rules or the second model handle routine cases. This partial application of complex modeling reduces the overall computational burden and device complexity while maintaining high accuracy where it matters most, avoiding excessive resource requirements across the entire system.
Data Source
AI summary
Sanitizing data can be a cumbersome task, particularly when the volume of data is large, the content is sensitive, and/or the type of sanitation requires contextual determinations. Sanitizing large amounts of data is tedious and may often require highly trained personnel with clearances and/or other qualifications. In the systems and methods of the present disclosure, language models (LMs) are used to solve these and other technical issues with tools that may allow sanitizing data easily, with high versatility, context awareness, and/or low demand for computational resources. In particular, some of the disclosed systems and methods use a first language model and a second language model (being less resource-intensive than the first language model) to generate sanitized output data with improved efficiency and accuracy. This dual-model approach ensures that sensitive information is handled appropriately while optimizing computer resource usage.