Differential privacy heterogeneous multi-attribute data publishing method based on vertical segmentation

Abstract

The invention discloses a differential privacy heterogeneous multi-attribute data publishing method based on vertical segmentation. A data owner Pi calculates a privacy budget epsilon i according to data sensitivity and privacy attribute exposure probability of a local data set Di; the data owner Pi constructs a hidden tree structure by using a hidden tree model learning algorithm, so that the constructed T meets epsilon < i1 >-differential privacy, and meanwhile, T calculated based on the T meets epsilon < i2 >-differential privacy; the Pi sends T and theta to a third party, and the third party constructs a hidden tree structure T of the integrated data set through a hidden tree model learning algorithm and updates a parameter theta at the same time; and according to the implicit tree structure T and the plus noise theta, sampling conditional distribution of each attribute, generating a disturbance data set D ', and issuing the disturbance data set D' to the outside. According to the invention, ideal data utility can be provided with lower communication and calculation costs.

Description

technical field [0001] The invention belongs to the field of computer software, and in particular relates to a method for issuing differentially private heterogeneous multi-attribute data based on vertical segmentation. Background technique [0002] The rapid development of network information technology has promoted the gradual popularization of Internet applications. While providing convenience, a large amount of data information is continuously generated by individual users, business units, research institutions, etc., and collected by various smart devices. By analyzing and processing the collected data, its potential value can be further tapped to create huge social benefits. In real life, a large amount of data is often collected from different smart devices, that is, the data of different attribute sets of the same individual is collected from different organizations or institutions, so that the data of a single individual is divided vertically by multiple parties. M...

AI Technical Summary

Problems solved by technology

[0005] The above methods deal with distributed settings with fixed privacy budget allocation, that is, to provide the same level of privacy protection for all data owners, which is obviously unreasonable

When multiple data owners jointly publish data, due to their own privacy risk control considerations, the privacy needs of all parties must be different. The same intensity of privacy protection will lead to insufficient protection of highly sensitive data owners, or Excessive loss of data information of the owner of low-sensitive data

Images