中文摘要：

针对目前存在大批网络用户,以群体形式来欺诈点击的问题,提出了一种检测点击欺诈群组的方法。首先使用频繁项集挖掘算法来发现共同点击过大量广告的个体用户,作为疑似欺诈组。然后,在对组内用户点击行为属性分析的基础上,运用孤立点检测方法找到与组内其他用户有显著差异的疑似欺诈用户。最后,运用贝叶斯分类方法对检测到的所有疑似欺诈成员分类,得到真正的欺诈群组和欺诈用户。在真实的数据集上进行的实验,验证了该方法的可行性和有效性。结果表明,该方法为点击欺诈检测问题提供了一条新的思路。

英文摘要：

This paper proposed a promising new method for detecting a fraudulent group. First,it used the frequent itemsets mining algorithm to reveal individual member with joint actions on advertisement click as a suspicious group. Next,based on properties analysis on click behavior of members of the suspicious group,it used outlier detection method to distinguish those whose behavior was significantly different with others,as suspicious targets. Finally,it detected the real fraud group and fraudulent users after applying Bayesian classifier on all the suspicious fraudulent users. Experimental results indicate the feasibility and validity of this method,which is based on real dataset. It gives a new approach to the detection of fraudulent clicks.