Privacy-preserving association rule mining in cloud computing

Recently, the paradigm of data mining-as-a-service in cloud computing environment has been attracting interests. In this paradigm, a company (data owner), lacking data storage, computational resources and expertise, stores its data in the cloud and outsources its mining tasks to the cloud service provider (server). In order to protect the privacy of the outsourced database and the association rules mined, k-anonymity, k-support, and k-privacy techniques have been proposed to perturb the data before it is uploaded to the server. These techniques are computationally expensive. If the data owner has resources to use these techniques, then it is often able to execute association rule mining locally. In this paper, we consider a scenario where a user (data owner) encrypts its data and stores it in the cloud. To mine association rules from its data, the user outsources the task to n (≥ 2) "semi-honest" servers, which cooperate to perform association rule mining on the encrypted data in the cloud and return encrypted association rules to the user. In this setting, we provide three solutions to protecting data privacy during association rule mining. Our solutions are built on the distributed ElGamal cryptosystem and achieve item privacy, transaction privacy and database privacy, respectively, as long as at least one out of the n servers is honest. To reduce the possibility that all servers are compromised, the user can use servers from different cloud providers. Our implementation and experiments demonstrate that our solutions are practical.