An Efficient Approximate Protocol for Privacy-Preserving Association Rule Mining

The secure scalar product (or dot product) is one of the most used sub-protocols in privacy-preserving data mining. Indeed, the dot product is probably the most common sub-protocol used. As such, a lot of attention has been focused on coming up with secure protocols for computing it. However, an inherent problem with these protocols is the extremely high computation cost – especially when the dot product needs to be carried out over large vectors. This is quite common in vertically partitioned data, and is a real problem. In this paper, we present ways to efficiently compute the approximate dot product. We implement the dot product protocol and demonstrate the quality of the approximation. Our dot product protocol can be used to securely and efficiently compute association rules from data vertically partitioned between two parties.