Study on Privacy Preserving Data Mining

调查问卷私密数据库的数据挖掘研究我们从基于网络的调查问卷中研究了私密评估的泄露问题，明确了泄露的潜在根源是从事公共服务器管理任务的恶意内部人员。为了防止恶意方的不诚实行为，我们应用了一种新的基于同态公钥算法的安全函数求值密码协议。然而，传统的安全功能评估协议不能在这里应用，因为在目前应用的问卷调查中使用了除了简单的n中选择类型之外的多种问题类型。为此，我们开发了一种基于析取式零知识证明协议的新协议，以克服问题风格复杂的问题。该方案在J.Nakazato和H.Kikuchi的《安全增强防止基于Web的调查问卷中的个人信息泄露》中进行了性能和安全性的研究，IPSJ期刊，第46卷，第8期，第2068-2077页，2005。我们还研究了其他数据挖掘算法，包括基于熵的决策树学习、基于欧几里德距离的聚类方案和不确定推理。一种应用是具有主观不确定性的自动网页分类系统。其中一些方案可以用来给出一种在不泄露密钥的情况下提取水印信息的新协议。我们已经展示了这些协议的准确性和性能等几个问题。

Study on Data Mining from Private Database of Questionnaires.We have studied the issue of disclosures of private evaluation from web-based questionnaire and clarified that the potential souse of disclosures is malicious insiders who engaged into administrative tasks of public server. In order to prevent dishonest behavior of malicious parties, we have applied a new cryptographical protocol of secure function evaluation using homomorphic public-key algorithms. The conventional protocol of secure function evaluation, however, can not apply here because a variety of question styles other than a simple one-out-of-n style are used in currently applied questionnaire survey. Thus, we have developed a new protocol based on disjunctive zero-knowledge proof protocol to overcome the issue of complicated style of questions. The performance and security of the proposed scheme are investigated in J.Nakazato and H.Kikuchi, Security Enhancement Preventing Personal Information Disclosure in Web-Based Questionnaire, IPSJ Journal, Vol.46, No.8, pp.2068-2077, 2005.We have also studied other data-mining algorithm including decision tree learning based on entropy, clustering schemes using Euclid distance and uncertain reasoning. One application is to automated web page classification systems with subjective uncertainty. Some of these schemes can be used to give a new protocol of extracting watermark information without revealing secret keys. We have shown several issues such as accuracy and performance of these protocols.