Privacy protection clustering method based on differential privacy

Abstract

A privacy protection clustering method based on differential privacy comprises the steps of: (1) enabling a data owner to calculate the Euclidean distance between data points in an original data set and construct a distance matrix; (2) adding differential noise to the distance matrix to form a noise-added distance matrix, and sharing the noise-added distance matrix to an untrusted data mining party to avoid data privacy leakage in a clustering analysis process; (3) enabling the data mining party to select k non-outliers with good global distribution from the noise adding distance matrix as initial center points of clustering; (4) calculating a sequence of q nearest central points of each data point, and distributing the data points to an expected interval of the q nearest central points to form a current round of clustering division; and (5) selecting the distance between the points in each cluster and the minimum data point as a new center point. The process is repeated until each cluster center point is not changed any more. According to the method provided by the invention, the clustering mining precision can be effectively improved while differential privacy protection of sensitive data is realized.

Description

technical field [0001] The present invention relates to the technical field of privacy protection, in particular to a privacy protection clustering method based on differential privacy. Background technique [0002] Data mining can discover potential regular patterns hidden in data and provide support for auxiliary decision-making. Clustering is the basic operation of data mining, and the access to individual data during the clustering process has the risk of leaking data privacy. For example, a pharmaceutical company hopes to cluster users' purchase records to obtain drug audiences, but access to sensitive drug purchase records such as skin diseases during clustering will violate patients' privacy. How to implement clustering mining under the premise of taking data privacy into account has become an urgent problem to be solved. Privacy-preserving clustering is an effective method to solve this problem, which has attracted continuous attention of researchers in recent years...

AI Technical Summary

Problems solved by technology

The deficiency of the prior art document 1 is that, after adding noise to the data object divided into each center point, the method of re-updating the center of the cluster is used to iteratively divide, resulting in the randomness of adding noise to the data point, and it is difficult to obtain the data from the data set. The impact of noise on data availability is considered from the perspective of overall distribution, resulting in insufficient clustering quality

Images