EsPRESSO: Efficient privacy-preserving evaluation of sample set similarity