Research On Privacy Protection Algorithm For Association Rules Mining

Data mining technology can efficiently use information resources.Association rule mining is one of the most important tasks in data mining and is widely used in marketing,securities trading,medical diagnosis and other fields.However,when mining data contain sensitive and private information,if no protection measures have been taken,information will be leaked.The traditional anonymous protection algorithm will generate security problems,and the encryption-based privacy protection algorithm lacks practicality.Therefore,there is an urgent need for a privacy protection algorithm with high availability and security to protect the privacy of association rules mining.Unlike traditional anonymous protection,differential privacy does not rely on the attacker's background knowledge,providing a quantifiable means of operation.Frequent pattern mining is the core of association rules mining.However,the existing differentially private frequent pattern mining has the problems of large amount of noise added,low availability,and low accuracy of mining results.For solving those obstacles illustrated above,the following two aspects are mainly concentrated on in this thesis.(1)A differentially private frequent itemset mining algorithm DP-FIM is proposed for privacy protection of frequent itemset mining.The existing algorithms have great damage to the data.In order to solve this problem,DP-FIM is based on the real frequent itemset to mine the frequent itemset with noise,so that the support of frequent itemset will not decrease,and the availability of mining results will be improved.Aiming at the problem of inconsistency between the support with noise and the real support for mining in existing algorithms,a consistent constraint strategy is proposed,which makes the mining support with noise in descending order of integer and ultimately improves the accuracy of mining results.Finally,theoretical analysis is used to prove the differential privacy and availability of the algorithm.The experimental comparison proves the high availability and high accuracy of the algorithm.(2)A differentially private frequent sequence mining algorithm DP-FSM is proposed for privacy protection of frequent sequence mining.In order to solve the problem that the existing algorithm has has excessively large amount of noise when mining long sequences,an optimal sequence length restriction strategy is applied in DP-FSM.By limiting the length of the sequence,the amount of noised added is reduced,which can effectively increase the availability of mining results.Aiming at the irrational use of the existing algorithm's privacy budget,an equivalent sequence merging strategy and a predictive support method are used to indirectly reduce the amount of added noise and improve the usability of the final result.In addition,the accuracy of mining results is improved by performing consistent constraint strategy on the support with noise.Finally,theoretical analysis is used to prove the differential privacy of the algorithm.The experimental comparison proves the high availability and high accuracy of the algorithm.