Key-Value Data Collection With Association-Preserving Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing local differential privacy technologies are inadequate for protecting the association between keys and values in key-value pair data, leading to reduced validity of statistical results when perturbing this data type.
Innovation Solution
A method and apparatus for collecting key-value pair data that ensures the association between keys and values by randomly setting the value of perturbed key-value pairs to either a or b based on a preset probability, using a differential privacy model to perturb the data and determine its inclusion in statistical results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing local differential privacy technology is used to perturb key-value pair data, then data privacy is protected, but the association relationship between keys and values is lost
Solution Approach 1:
The patent merges the perturbation processes of keys and values into a unified operation. When a key is perturbed to a new key, the corresponding value is also perturbed to maintain their association. This combining approach ensures that the differential privacy guarantee holds for the key-value pair as a unit, preventing the loss of association relationships while still protecting individual privacy.
Solution Approach 2:
The patent introduces an intermediary mechanism that tracks and maintains the mapping between original keys and perturbed keys. This intermediary structure allows the system to preserve the association between keys and values even after perturbation by using the intermediary to resolve the correspondence during statistical aggregation, thus preventing information loss while maintaining privacy.
2Reliability
If key-value pair data is perturbed to protect privacy, then individual data points are protected, but the validity of statistical results is reduced
Solution Approach 1:
The patent changes the parameter representation by treating key-value pairs as unified entities rather than separate elements. By modifying the perturbation parameters to work on pairs simultaneously and introducing new parameters for tracking associations, the system maintains statistical validity while protecting individual data points. The parameter changes enable the statistical aggregation to account for the perturbed relationships correctly.
3Reliability
If traditional perturbation methods are applied separately to keys and values, then each element is protected, but the association relationship is broken
Solution Approach 1:
The patent merges the perturbation operations for keys and values into a single coordinated process. Instead of applying independent perturbation functions to each element, the system applies a joint perturbation mechanism that ensures when a key changes, its associated value changes in a corresponding manner, thereby maintaining the stability of their association relationship while still providing element-level privacy protection.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
This application provides a method and an apparatus for sending key-value pair data, and a method and an apparatus for collecting key-value pair data. In the method, first target key-value pair data is perturbed as first perturbed key-value pair data, where a value of the first perturbed key-value pair data is randomly set to 〈a,v1*〉 or 〈b,v〉 based on a preset probability, so that a data collector can determine, based on a value a or b of a first value in the first perturbed key-value pair data, whether to include the entire first perturbed key-value pair data into the statistical result in a process of generating the statistical result. In this way, an association between a value and a key in the key-value pair data is reserved in the first perturbed key-value pair data, which improves availability of the first perturbed key-value pair data. That is, a statistical result generated by the data collector based on the first perturbed key-value pair data is closer to a statistical result generated based on original user key-value pair data.