基于多元特征的微博被劫持账户检测

Compromised Accounts Detection in Weibo Based on Multiple Features

微博作为一个拥有大量用户的社交平台,其较大的影响力与开放性吸引了攻击者的目光。攻击者会利用异常账户进行网络犯罪,对异常账户的检测是维护社交网络安全的重要途径之一。基于攻击者的常规行为以及正常用户的行为特征,提出了一种微博中被劫持账户的检测方法。该方法使用6个特征对用户行为进行分析,使用异常得分刻画用户行为的异常程度,使用传统机器学习分类器检测被劫持账户。为验证本文方法的有效性,采用了由复旦大学提供的公开微博数据集进行实验,结果表明,采用决策树分类器时,本文方法对被劫持账户的检测精确率高达97.5%。

As a social platform with a large number of users, Weibo has attracted the attention of attackers because of its significant influence and openness. Attackers will utilize anomaly accounts for cybercrime. One of preserving social network security methods is detecting anomaly accounts. Based on attackers’ routine behavior and benign users’ behavior features, this paper proposes a method to detect compromised accounts. In this paper, we use six features to analyze users’ behaviors and propose the abnormal scores to evaluate the degree of violation. To verify the effectiveness of our method, the public Weibo dataset provided by Fudan University is used in the experiment and the results show that the precision of our method for detecting compromised accounts can achieve 97.5% when using the decision tree classifier.