摘要
针对社交网络内容安全,存在一种通过属性推断补全来获取用户私密属性的攻击。传统的基于无监督学习的方法和基于监督学习的属性补全攻击方法存在未能把结构相似性和同质性有效结合起来的问题。文章提出了一种基于隐式表达的用户属性补全攻击方法,把用户属性补全抽象为一个有监督的分类问题,基本思路是利用node2vec算法将社交网络中的用户节点映射成向量,然后将向量通过聚类方法计算一个节点所在的社区,在社区内构建分类模型,并利用此模型对用户缺失属性进行预测。文章在真实数据集上进行验证,证明了算法能够有效提高社交网络用户属性补全的准确率。
In social networks, there is an attack threatening its content security by acquiring user private attributes from attribute inference completion. Traditional user attribute completion methods like unsupervised learning and supervised learning fail to effectively combine homogeneity with structural similarity. This paper presents a user attribute completion attacking method based on implicit expression, which abstracts user attribute completion as a supervised classification problem. The basic idea is to use node2vec algorithm to map the user nodes in social networks into vectors, and then use the clustering method to calculate the community where a node is located, construct the classification model in the community, and use this model to predict the missing attributes of the user. This paper verifies that this algorithm can improve the accuracy of user attribute completion in social networks on a real data set.
出处
《信息网络安全》
CSCD
2017年第12期67-72,共6页
Netinfo Security
基金
国家重点研发计划[2016YFB0800504]