Research On Privacy-Preserving Data Mining Of Classification In Cloud Storage

With the rapid development of cloud computing technology and the explosive growth of human information,more and more enterprises and individuals attempt to store their data in the cloud.However,the expansion of the cloud storage business has also brought data security problems to users.Faced with the vast data on the cloud,how to mine data effectively under the case of protecting user data privacy,which can make people obtain some innovative and valuable information,has become an important topic research.Homomorphic encryption technology has unique advantages in ciphertext processing,which can effectively solve the conflict between data privacy protection and information reprocessing.This paper is based on the research of privacy preserving data mining combined with homomorphic cryptography,the works of this paper are as follows:1.We propose a privacy-preserving profile-matching scheme over improved HRES algorithm in mobile social network.Due to the transparency of the wireless channel,users in multiple-keys environment are vulnerable to eavesdropping during the process of uploading personal data and re-encrypting keys,which also imposes additional key management burden.In addition,profile matching using vector inner product cannot effectively filter users with ulterior motives.To tackle above challenges,an improved HRES homomorphic re-encryption algorithm is proposed in the scheme,so that it can satisfy the extra single multiplication homomorphism while satisfying any addition homomorphism.The public key negotiated by the clouds is used to encrypt the users' data,thereby avoiding key leakage and key management issues,and the security of the system is improved.Furthermore,the scheme utilizes the homomorphic multiplication property of the improved HRES algorithm,and adopts the cosine result between the normalized vectors as the standard for measuring the users' proximity,which can effectively improve the social experience of the users.2.We propose an outsourced privacy-preserving classification scheme over encrypted data.The classifier owner outsources the encrypted classifier to the cloud server,and entrusts the cloud as an agent to provide the user with remote classification service.The classification service are accompanied by low computational efficiency and high communication cost,and malicious users can attack the protocol by replacing encryptedvalues to be compared to get relevant information.In order to avoid the above drawbacks,we use a more efficient OU algorithm as the main encryption scheme of the protocol,which effectively reduces the computational cost and communication cost.In addition,for a simple malicious adversary model,we have made further modification to the scheme so that malicious users can only obtain the confused encrypted values,which will not affect the comparison results.At the end,we proved that our scheme is secure and efficient through the security analysis and experimental simulation.