一种基于用户行为特征选择的点击欺诈检测方法

Click Fraud Detection Method Based on User Behavior Feature Selection

在线广告是目前众多网络巨头收入的主要来源,在线广告也为网络的健康发展提供了强大的经济支撑。目前,利用用户行为属性特征来识别点击欺诈的方法中,含有较多的冗余特征,检测效率相对较低。针对这一问题,提出了一种属性特征选择与分类方法相结合的欺诈检测方法。通过训练数据集找到欺诈用户点击广告的属性特征集合,采用Fisher分方法得到了属性特征重要度排序,选取重要属性特征,并基于这些重要的特征使用支持向量机二分类方法分类。在真实数据集上的实验结果证明了该方法的可行性与有效性。

Online advertisement is not only the main sources of income of profit for internet giants, but also provides powerful economic support for the internet development. The commonly used methods of click fraud detection, which are based on the features of client＇s behavior,may lead to inefficiency in fraud detection due to redundant features. To solve this problem, a fraud detection method which combines feature selection with classification method was proposed. According to the feature attributes set of fraud advertisement which is found through training set, attribute significance is sorted by Fisher score method. The important attributes is selected and the SVM algorithm is lastly introduced into classification based on these important attributes. Experiments on real data set demonstrate that the proposed detection method is feasible and valid.