Privacy-preserving similarity coefficients for binary data

Similarity coefficients (also known as coefficients of association) are important measurement techniques used to quantify the extent to which objects resemble one another. Due to privacy concerns, the data owner might not want to participate in any similarity measurement if the original dataset will be revealed or could be derived from the final output. There are many different measurements used for numerical, structural and binary data. In this paper, we particularly consider the computation of similarity coefficients for binary data. A large number of studies related to similarity coefficients have been performed. Our objective in this paper is not to design a specific similarity coefficient. Rather, we are demonstrating how to compute similarity coefficients in a secure and privacy preserved environment. In our protocol, a client and a server jointly participate in the computation. At the end of the protocol, the client will obtain all summation variables needed for the computation while the server learns nothing. We incorporate cryptographic methods in our protocol to protect the original dataset and all other intermediate results. Note that our protocol also supports dissimilarity coefficients.