社交网络用户敏感属性迭代识别方法

Sensitive attribute iterative inference method for social network users

分析识别社交网络用户敏感信息,有利于从技术上量化隐私泄露程度,进行隐私保护。针对现有的用户属性识别方法需要对用户属性取值进行强假设的问题,结合RL迭代分类框架和扩展wvRN关系识别的方法,提出了一种社交网络用户敏感属性迭代识别方法。通过卷积神经网络提取用户文本特征进行识别,结合邻居结点迭代地进行关系识别,不仅弱化了对用户属性的假设,而且提高了可用性。实验结果表明,通过在社交网络中获取少量的标注数据,对迭代识别方法设置合理的参数值,可以获得较好的用户敏感属性识别结果。

Analyzing and inferring sensitive information of social network users is conducive to technically quantifying the degree of privacy leakage and protecting privacy. Aiming at the problem that existing user attribute inference methods needs to make strong assumptions on the value of user attributes, an iterative method for user sensitive attributes in social network is proposed by combining the RL iterative classification framework and extending the wvRN relation inference method. Extracting probabilities of user sensitive attributes based on user text and convolution neural network and iteratively updating inference results with neighboring nodes, not only weakens the assumption of user attributes, but also improves the degree of application. The experimental results show that by obtaining a small amount of labeled data in social networks and setting reasonable parameter values for iterative inference methods, better user sensitive attribute inference results can be obtained.