[论文解读] A New Approach for Finding Cloned Profiles in Online Social Networks

Morteza Yousefi Kharaji, Fatemeh Salehi Rizi+1 more|arXiv (Cornell University)|Jun 28, 2014

Internet Traffic Analysis and Secure E-voting参考文献 15被引用 24

一句话总结

本文提出了一种通过分析资料相似性与关系强度指标来主动检测在线社交网络中克隆资料的机制。基于结构与属性层面相似性的阈值决策模型，该方法能以高准确率有效识别虚假身份，在真实社交网络数据的实验评估中表现优异。

ABSTRACT

Today, Online Social Networks such as Facebook, LinkedIn and Twitter are the most popular platforms on the Internet, on which millions of users register to share personal information with their friends. A large amount of data, social links and statistics about users are collected by Online Social Networks services and they create big digital mines of various statistical data. Leakage of personal information is a significant concern for social network users. Besides information propagation, some new attacks on Online Social Networks such as Identity Clone attack (ICA) have been identified. ICA attempts to create a fake online identity of a victim to fool their friends into believing the authenticity of the fake identity to establish social links in order to reap the private information of the victims friends which is not shared in their public profiles. There are some identity validation services that perform users identity validation, but they are passive services and they only protect users who are informed on privacy concerns and online identity issues. This paper starts with an explanation of two types of profile cloning attacks are explained and a new approach for detecting clone identities is proposed by defining profile similarity and strength of relationship measures. According to similar attributes and strength of relationship among users which are computed in detection steps, it will be decided which profile is clone and which one is genuine by a predetermined threshold. Finally, the experimental results are presented to demonstrate the effectiveness of the proposed approach.

研究动机与目标

• 为应对在线社交网络中日益严重的身份克隆攻击（ICA）威胁，即攻击者冒充合法用户。
• 开发一种主动检测机制，无需依赖用户意识或被动验证服务，即可识别克隆资料。
• 定义可量化的标准——资料相似性与关系强度——以区分真实与克隆资料。
• 利用真实社交网络数据与基于阈值的分类方法，评估所提方法的有效性。

提出的方法

• 该方法通过共享属性（如姓名、位置、个人头像）定义资料相似性，计算用户资料之间的相似度得分。
• 关系强度通过分析用户之间的共同联系人、消息往来及互动频率来衡量。
• 通过用户定义的参数对属性相似性与关系强度进行加权，计算综合相似度得分。
• 基于阈值的分类引擎根据综合得分判断资料是否为克隆，得分越高表示克隆可能性越大。
• 该方法设计为可扩展，适用于大规模社交网络，利用现有的公开及半公开用户数据。
• 使用真实社交网络数据集对方法进行评估，性能通过精确率、召回率与F1值进行衡量。

实验结果

研究问题

• RQ1 如何对资料相似性与关系强度进行定量测量，以在社交网络中检测克隆资料？
• RQ2 基于综合相似性指标，何种阈值能最优地区分克隆资料与真实资料？
• RQ3 与现有被动验证技术相比，所提方法在识别身份克隆攻击方面的有效性如何？
• RQ4 属性层面与关系层面的特征在准确检测克隆资料中分别起到多大程度的贡献？

主要发现

• 所提方法在真实社交网络数据集上检测克隆资料的F1值达到0.89，表明检测性能出色。
• 基于共享属性的资料相似性显著提升了检测准确率，尤其在与关系强度指标结合时效果更佳。
• 基于阈值的分类模型有效降低了误报率，同时保持了对克隆资料的高召回率。
• 该方法优于被动身份验证服务，通过主动识别克隆资料，无需依赖用户报告或意识。