Research On Frequent Itemset Mining Based On Supermarket Transactions

With the development of information technology and the rise of cloud services,in order to store and manage massive data,many supermarkets upload customers' transactions to the Cloud Service Provider(CSP).Therefore,many transactions are stored on the CSP.What's more,the CSP has powerful computation ability.Hence,data mining can be executed with the help of the CSP.According to the mining result,supermarkets can arrange goods appropriately,which can not only provide convenience to customers but also improve supermarkets' profit.Because association rules mining is one of the most significant methods among different data mining methods,and frequent itemset mining is the fundamental step of association rules mining,it is necessary to introduce a secure and efficient frequent itemset mining method.The method is required to output a correct mining result,have high efficiency and provide users' privacy from leaking during the process of mining.The main work of this thesis is listed as follows:1.We introduce a frequent itemset mining method against inner attacks.In order to protect users' privacy,some methods are designed based on cryptology.Users encrypt transactions using Evaluator's public key and then upload the ciphertexts to the CSP.As it cannot resist inner attacks executed by the Evaluator,we change the basic cryptosystem to BCP cryptosystem in our method.Transactions are encrypted using the joint public key instead of the Evaluator's public key.Then the ciphertexts are uploaded to the CSP.Hence,even if the ciphertexts are intercepted by the Evaluator during this process,the Evaluator cannot decrypt and obtain the original transactions because it does not have the private key.Therefore,this method can prevent the problem of users' privacy leakage.Besides,when calculating scalar products,because the mining query is a binary vector,we mark the items with the value of 1 in the mining query.Based on homomorphic properties,we can obtain the ciphertexts of scalar products by performing multiplications on the marked items in each encrypted transaction record,rather than calculating all items,which can help decrease redundant calculation.Furthermore,we improve the process of randomization,which will help get correct evaluation results after evaluating encrypted scalar products under BCP cryptosystem.In conclusion,our method can not only output a correct mining result,but also have high-level efficiency and security during the process of mining.2.We introduce a frequent itemset mining method under the encrypted mining query.In the exiting methods,although the mining can be operated for the encrypted mining query,it needs a large amount of calculation and has low efficiency.In order to solve this problem,we design a blocking algorithm based on the sparse property of the transactions.The encrypted transactions and the encrypted mining query are split using the blocking algorithm.During the process of calculating the encrypted scalar products,the bilinear pairings are only needed to be calculated on the ciphertexts in parts of blocks,rather than all ciphertexts.Therefore,the number of bilinear pairings that are needed to be calculated is cut down,the running time is decreased,and the mining efficiency is enhanced.Besides,we also improve the process of randomization,which will help get correct evaluation results after evaluating the mixed blocks and the encrypted scalar products under BGN cryptosystem.In conclusion,our method can not only output a correct mining result,but also have high-level efficiency and security during the process of mining.