保护隐私的集合相似性度量协同计算协议

Privacy Preserving Set Similarity Measurement Collaborative Computing Protocol

集合相似性度量是机器学习领域的基本问题之一,研究如何在保护数据隐私的前提下计算两个集合间的相似性问题,在保护数据隐私的机器学习、图形识别、生物信息学等方面有着重要的理论意义与应用价值。在机器学习中估算不同样本集合之间的相似性时,通常通过计算集合相似度来对样本之间的相似程度进行估算,这一类集合之间的相似度统称为集合距离。其中,最常用到的集合距离就是杰卡德距离。文中从集合间杰卡德距离入手,首先通过设计一种新的编码方法,对参与计算的数据进行位置数字编码,将相似性度量问题转化为求两集合间相同数字个数问题,进而结合异或思想,借助同态加密体制具体设计了可以保护隐私的集合杰卡德距离协同计算协议,从而解决了集合间相似性度量的隐私保护问题。模拟器证明该协议是安全的,结果分析表明协议可以高效安全地判定出两对象间集合数据的相似性,在保护隐私的集合相似性度量方面,该方法具备一定的普适性。

Set similarity measurement is one of the basic problems in the field of machine learning. Studying how to calculate the similarity between two sets on the premise of protecting data privacy has important theoretical significance and application value in machine learning, graphics recognition, bioinformatics and so on. When estimating the similarity between different sample sets in machine learning, the similarity degree between samples is usually estimated by calculating the set similarity. This kind of similarity between sets is collectively referred to as set distance. Among them, the most commonly used set distance is Jaccard distance. Starting with the Jaccard distance between sets, we firstly design a new coding method to encode the position numbers of the data involved in the calculation, transform the similarity measurement problem into the problem of finding the number of the same numbers between two sets, and then design a set Jaccard distance collaborative calculation protocol that can protect privacy with the help of homomorphic encryption system, Thus, the privacy protection problem of similarity measurement between sets is solved. The simulator proves that the protocol is secure. The result analysis shows that the protocol can effectively and safely determine the similarity of set data between two objects. This method has certain universality in the measurement of set similarity to protect privacy.