摘要：

传统利用假位置生成技术构建匿名集的隐私保护方案未进行综合因素的考量，致使最终生成的假位置区分度低、合理性差，存在较高的隐私泄露风险。提出基于多属性决策模型的匿名集构建（MDMASC）算法，基于多属性决策模型选取假位置，从而构建安全匿名集。考虑到攻击者可能具有背景知识等信息，通过对地图进行网格划分，根据用户对各位置点的历史查询概率进行初次过滤。考虑到匿名集语义多样性、物理分散性、假位置敏感程度等因素，定义语义跨度、语义敏感等级、位置普遍度等指标属性。利用层次分析法建立层次结构模型，分析位置普遍度、语义敏感等级、语义跨度等5个影响匿名集安全性的指标之间的相对重要关系，基于此构建成对比较矩阵，计算出各指标的属性权重。最后，利用多属性决策模型计算候选假位置集中各位置点的综合属性值，选出最优假位置构建安全匿名集。实验结果表明，MDMASC算法相对于MMDS算法降低了约10.3%时间开销，被语义攻击算法识别的概率相比MMDS算法和K-DLS算法分别降低了14.9%和25.5%，在满足用户隐私要求的前提下具备可行性和有效性。

关键词: 位置隐私, 多属性决策, 层次分析法, 语义信息, 匿名集

Abstract:

The traditional privacy protection scheme for constructing anonymous sets using dummy location generation technology does not consider comprehensive factors. This results in low differentiation and poor rationality of the final generated dummy locations, thus posing a high privacy leakage risk. Therefore, Multi-attribute Decision Model-based Anonymous Set Construction(MDMASC) is proposed in this study, which selects dummy locations based on a multi-attribute decision model and subsequently constructs secure anonymous sets. Considering that an attacker may possess background knowledge and other information, the algorithm first grids the map and performs initial filtering based on the probability of the users' historical queries for each location point. Second, considering the semantic diversity, physical dispersion and sensitivity of the anonymous set dummy location, it defines index attributes such as the semantic span and sensitivity level, and the location prevalence. Subsequently, hierarchical analysis is used to establish a hierarchical structure model, and a pairwise comparison matrix is constructed by comparing the relative importance relationships among five indicators affecting the security of anonymous sets, such as location prevalence and semantic sensitivity level. Next, each indicator's attribute weight is calculated, after which, a multi-attribute decision model is utilized to calculate the comprehensive attribute values of each location point in the candidate dummy location set.Finally, the optimal dummy location is selected to build a secure anonymity set. Experimental results demonstrate that the proposed MDMASC algorithm reduces the time cost by approximately 10.3% compared with the MMDS algorithm, and the probability of being recognized by semantic attack algorithms decreases by 14.9 and 25.5% compared with the MMDS and K-DLS algorithms, respectively, which is considered feasible and effective.

Key words: location privacy, multi-attribute decision-making, Analytic Hierarchy Process(AHP), semantic information, anonymity set