在不同再識別風險下基於資訊遺失的去識別化策略探索

Abstract

對於資料分析的研究來說，尤其是醫療資訊方面，資料個體本身的 隱私性以及資料可應用程度是一個難解的取捨。基於對於個人隱私的 需求，在我們發布資料前會以盡量保持資料的完整性為前提將其進行 某種程度的模糊化，讓惡意使用者無法分辨其真實的個體。此稱為去 識別化技術。 此篇論文提出了在不同的去識別化風險下快速有效率地找到資料完整 性較佳的去識別化方式，藉由將問題本身映射到一個晶格模型(lattice model) 的探索問題上，並且使用了子晶格(sublattice) 的概念加速我們 的搜尋。實驗結果顯示此種方法在同樣的搜索數目下，可以找到資訊 遺失量較低的去識別化方式。

In the study of data analysis, especially in the field of medical information, there is a complex tradeoff between the privacy of the data and the applicability of the data. Due to the demand for individual privacy on maintaining the integrity of the original data, the data are being more or less obscured before the release to avoid the malicious users from identifying the true owners. This technique is called de-identification. This dissertation introduces a novel technique that is able to efficiently and quickly find a de-identification policy with satisfactory data integrity under variant re-identification risks. We probe into the problem by applying it onto the lattice model [12] and use the concept of sublattice [13] to accelerate the process. The results appear to be promising as in comparison to other techniques [12], we can find de-identification policies with less informationloss under the same search times.