Reseach On Algorithms For Mining Association Rules Satisfied With Differential Privacy

Association rule mining, which is an important research field of data mining, have been widely applied in field such as salling, medical etc. Its purpose is to find out the correlations between different items contained in dataset. When privacy information contained in dataset, association rule mining may leak the user’s privacy. So, it is necessary to protect these privacies in association rule mining.The differential privacy protection model has better privacy protection effect when compared with the traditional privacy protection model. Its mining results owns a better usability and effective privacy protection at the same time. Two modified algorithms of association rule mining, LMS and EMS, that satisfied differential privacy are proposed to protection the user’s privacy information in the dataset against the propogation problems existed in algorithm SmartTrunc. For algorithm LMS, the Laplace noise is used to interfere the candidate itemsets’ true support counts, and the frequent itemsets are acquired according to the disturbed support count values of the candidate itemsets. For algorithm EMS, both the exponential mechanism and the Laplace noise disturbance have been applied. Baed on the privacy protection theory discussed, some counting examples are shown. At last, the privacy and usability of these two modified methods have been analysed and verified. The results show that these two algorthms have good effect both in performance of privacy and accuracy.