距離及相關性資料探勘演算法的隱私權保護

Abstract

這篇論文設計出一個轉換方式使得當資料送到第三方被研究時還能保護到資料的隱私性。大部分傳統的轉換方式都有兩種限制，演算法侷限性與資訊量流失。在這篇論文中，我們提出了一個新穎的隱私權保護方式而沒有這兩種限制。這種轉換演算法我們稱之為FISIP: 一階和、二階和與內積維護。特別的是，我們將證明，藉由FISIP保護隱私資料的這三種性質（一階和、二階和與內積），當它被轉換成公開資料時，資料還能用在只依據這三種性質的所有演算法。由於距離與相關性能從這三種性質推導出來，因此，只依據距離與相關性的所有演算法依舊能被應用到。FISIP的評估有兩部分，第一部分是資料的有用性，第二部分是資料的強大性，這兩個目標本質上很難同時被達到。然而，從我們的實驗結果顯示，FISIP能同時滿足這兩個目標。總而言之，FISIP能提供一種轉換使得轉換前的原始資料與轉換後的公開資料的距離及相關性皆一致。當資料的隱私性被保護到時，資料的探勘品質在轉換後的（公開）資料能與轉換前的（隱私）資料達到一致。

This paper devises a transformation scheme to protect data privacy in the case that data has to be sent to the third party for analysis purpose. Most conventional transformation schemes suffer from two limits, i.e. algorithm dependency and information loss. In this paper, we propose a novel privacy preserving scheme without these two limitations. This transformation algorithm is referred to as FISIP: FIrst and Second order sum and Inner product Preservation. Explicitly, as will be proved, by preserving three basic properties, (i.e. first order sum, second sum, and inner products) of private data, algorithms whose measures can be derived from the three properties can still be applied to public data transformed by FISIP. Specifically, distance and correlation can be derived from the three properties. Hence, distance-based algorithms and correlation-based algorithms can be applied. Evaluation of FISIP is done in two parts. The first part is data usefulness. The second part is data robustness. The two goals are intrinsically difficult to achieve at the same time. However, FISIP attains these two goals shown by our experimental results later. In all, FISIP is able to provide a transformation that preserves the distance and the correlation for the original private data after their transformation to the public data. As a result, while the privacy is protected, the mining quality from the transformed (public) data can be obtained to be the same as that from the original (private) data.