Towards semantically secure outsourcing of association rule mining on categorical data

When outsourcing association rule mining to cloud, it is critical for data owners to protect both sensitive raw data and valuable mining results from being snooped at cloud servers. Previous solutions addressing this concern add random noise to the raw data and/or encrypt the raw data with a substitution mapping. However, these solutions do not provide semantic security; partial information about raw data or mining results can be potentially discovered by an adversary at cloud servers under a reasonable assumption that the adversary knows some plaintext–ciphertext pairs. In this paper, we propose the first semantically secure solution for outsourcing association rule mining with both data privacy and mining privacy. In our solution, we assume that the data is categorical. Additionally, our solution is sound, which enables data owners to verify whether there exists any false data in the mining results returned by a cloud server. Experimental study shows that our solution is feasible and efficient.