EsPRESSo: Efficient Privacy-Preserving Evaluation of Sample Set Similarity