一种引入反馈惩罚机制的个性化数据匿名发布模型

发布时间：2018-10-22 19:52

【摘要】：为了避免个人隐私信息被泄露,一般数据集在发布时都会进行匿名化脱敏处理,使得攻击者无法从发布的数据中找到具体个人的隐私信息,从而避免受到名誉、财产或身体方面的损失。在当前的信息时代,作为最常见的个人数据发布场景,基于个人信息的网络服务就不可避免的成为重灾区。传统的网络服务模型只存在服务方和通信网络提供的隐私保护,但当用户信息成为商家的掘金地时,用户的个人信息不免被恶意收集,从而使得用户成为“无秘之人”,随时会受到未知源头的恶意攻击。本文以隐私保护为切入点,探讨了交互式个人数据发布场景下的隐私信息保护及其效用平衡的问题。在综合分析了各种匿名保护发布原则后,提出了在传统随机博弈的理论框架下引入反馈惩罚机制,并用个性化属性泄露风险之和低于隐私泄露容忍度对博弈结果进行纠错的新想法,构造了一个引入反馈惩罚机制的个性化数据匿名发布模型,并用实验对其效果进行了验证。该模型基于服务方与用户之间的服务过程可以抽象为一个混合策略完全信息静态博弈,通过求解混合策略纳什均衡为用户选择最佳的应对策略,始终使用户获得最大博弈收益。实验证明该模型确实对前者进行了有效改善,结论主要体现在两点:1)文中所提出的模型对具体用户在数据效用率、隐私保护度、模型贡献率三方面具有稳定性,也即,此三者不会随着用户服务发起次数的改变而改变,这体现了模型本身的稳定性。2)该模型的使用效果与用户本身的个性化属性配置有关,不同用户能得到的数据效用率、隐私保护度是不同的,这一方面体现了模型的个性化,另一方面也能最大程度保证数据效用与隐私保护之间的平衡。
[Abstract]:In order to avoid the disclosure of personal privacy information, the general data set is desensitized anonymously when it is published, which makes it impossible for an attacker to find the privacy information of a specific individual from the published data, thus avoiding being reputed. Loss of property or body In the current information age, as the most common personal data release scenario, the network service based on personal information will inevitably become a disaster area. The traditional network service model has only the privacy protection provided by the service side and the communication network, but when the user information becomes the gold mine of the merchant, the personal information of the user is collected maliciously, which makes the user become the "unsecretive person". At any time will be the unknown source of malicious attacks. Based on privacy protection, this paper discusses privacy information protection and its utility balance in interactive personal data publishing scenarios. After a comprehensive analysis of various anonymous protection release principles, a feedback penalty mechanism is proposed under the framework of traditional stochastic game theory. Based on the new idea that the sum of the risk of personalized attribute leakage is lower than the tolerance of privacy disclosure, a new idea of correcting the result of game is proposed, and a model of anonymous publication of personalized data with feedback penalty mechanism is constructed, and its effect is verified by experiments. This model can be abstracted as a mixed strategy complete information static game based on the service process between the service party and the user. By solving the Nash equilibrium of the mixed strategy to select the best coping strategy for the user, the user can always obtain the maximum benefit of the game. Experimental results show that the model can effectively improve the former. The conclusions are as follows: 1) the proposed model is stable to specific users in three aspects: data utility rate, privacy protection, and model contribution rate, that is, the proposed model is stable in terms of data utility rate, privacy protection degree and model contribution rate. These three do not change with the number of user service initiation, which reflects the stability of the model itself. 2) the use of the model is related to the user's own personalized property configuration, different users can get the data utility rate. The degree of privacy protection is different, which reflects the individuation of the model, on the other hand, it can ensure the balance between data utility and privacy protection.
【学位授予单位】：湖北师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP309

【参考文献】