TC: Small: Collaborative: Protocols for Privacy-Preserving Scalable Record Matching and Ontology Alignment

许多应用领域，如情报、反恐、取证、疾病控制，往往需要交叉匹配多个非常大的数据集，如监视列表。由于这些数据集可能包含隐私敏感或机密信息，因此使用高效的隐私保护协议来交叉匹配不同的数据集至关重要。使用安全多方计算(SMC)协议解决了隐私保护记录匹配问题，在这些协议下，数据被转换为具有私有输入的一系列函数。然而，基于SMC的协议的一个主要缺点是它们涉及大量的密码原语，如同态加密，这些原语不能扩展到实际问题的大小。因此，基于SMC的协议不能直接用于资源受限的数据密集型隐私保护记录匹配方法。该项目开发了一种基于观察到的新方法，即要将SMC应用于实际应用，需要弥合使用SMC协议可以有效匹配的数据集的大小与实践中看到的数据集的大小之间的差距。该项目采取的方法从一个新的角度解决了这个问题，开发了一些技术，通过使用保护隐私的数据清理方法来缩小实际问题的规模。因此，该项目通过以下步骤解决了隐私保护数据匹配问题。首先，为了保护数据主体的隐私，使用差异隐私收集关于数据的有用统计数据。第二，参与数据匹配的各方共享不同的私人统计数据。然后，这些缔约方确定可能发生卓有成效的匹配的潜在匹配对。这样的步骤被称为数据阻塞。最后，将SMC技术应用于这些候选对象，以准确地交叉匹配信息。语义匹配协议除了支持句法匹配外，还支持根据一定的语义相似度函数对记录进行比较。语义匹配协议包括本体匹配和对齐技术，因为本体的使用是有效的语义匹配的关键。这个项目是第一个使用差异隐私来有效保护隐私的记录匹配，它还利用基于语义的方法和隐私保护方法来进行本体对齐。该项目开发的技术首次使用差异隐私实现了大规模数据集的高效隐私保护匹配，从而克服了传统SMC技术的可扩展性问题。该项目开发的方法通过在不侵犯数据隐私的情况下实现可能由不同各方拥有的多个数据档案的交叉匹配，扩大了数据使用的机会和背景。许多对我们的社会有意义的应用程序将从这些机会中受益。有关更多信息，请参阅项目网站URL:http://www.cs.purdue.edu/homes/bertino/prirelink

Many application domains, such as intelligence, counter-terrorism,forensics, disease control, often need to cross-match multiple verylarge datasets, such as watch lists. Because those datasets may contain privacy-sensitive or confidential information, the use of efficientprivacy-preserving protocols for cross-matching different datasets iscrucial. The problem of privacy-preserving record matching has beenaddressed by the use of Secure Multi-party Computation (SMC) protocols.Under these protocols, the data are converted to series of functionswith private inputs. However a major drawback of SMC-based protocolsis that they involve extensive cryptographic primitives such ashomomorphic encryption which do not scale to the size of practicalproblems. As a result, SMC-based protocols cannot be used for resourceconstrained data-intensive privacy-preserving record matching approaches directly. This project develops a novel approach based on the observationthat to apply SMC to practical applications, one needs to bridge the gap between the size of the datasets that can efficiently be matched using SMC protocols and the size of the datasets seen in practice. The approach taken by the project tackles the problem from a novel angle by developing techniques to reduce the size of practical problems by employing privacy-preserving data sanitization methods. The project thus solves the privacy-preserving data matching problems through the followingsteps. First, to protect the privacy of data subjects, useful statistics about data is gathered using differential privacy. Second, differentially private statistics are shared among the parties involved in data matching. These parties then identify potential matching pairs where fruitful matching may occur. Such a step is referred to as data blocking. Finally, SMC techniques are applied to these candidates to accurately cross-matchinformation. In addition to syntactic matching, semantic matching is supported by which records are compared according to some semantic similarity functions.The semantic matching protocols includes techniques for matching andaligning ontologies, as the use of ontologies is crucial for an effectivesemantic matching. This project is the first to use differential privacy forefficient privacy-preserving record matching that also leverages semantics-basedapproach and a privacy-preserving approach to ontology alignment. The techniquesdeveloped in the project are the first to achieve efficient privacy-preservingmatching of large scale data sets using differential privacy, thus overcoming the scalability problems of conventional SMC techniques. The approach developed in this project expands the opportunities and contexts for data use by enabling the cross-match of multiple data archives, possibly owned by different parties, without violating the privacy of the data. Many applications, of interest for our society, will benefit by such opportunities.For further information see the project web site at the URL:http://www.cs.purdue.edu/homes/bertino/prirelink