基于共生矩阵的电子商务交易日志异常检测  

Anomaly Detection on E-commerce Transactions Log based on Co-occurrence Matrix

作　　者：全拥[1] 李树栋[1,2] 贾焰[1] 韩伟红[1] 

机构地区：[1]国防科学技术大学计算机学院,湖南长沙410073 [2]山东工商学院数学与信息科学学院,山东烟台264005

出　　处：《中国电子商情（通信市场）》2013年第4期39-45,共7页

基　　金：国家高技术研究发展计划（2012AA01A401）;国家自然科学基金（61202362,61262057）

摘　　要：针对电子商务中用户交易行为合法与否的问题，提出了一种基于共生矩阵的异常检测算法。该算法利用共生矩阵对用户的交易行为建模，通过PCA方法建立共生矩阵空间，从而得到用户正常交易模式。在检测阶段，对待测数据产生的共生矩阵进行了修正并获取用户的交易模式，通过矩阵2一范数计算用户交易模式和其正常模式之间的距离并以此来判断用户的交易行为是否异常。实验表明，相比于其它的几种方法，本文的方法具有更高的检测性能。In order to determine whether the user behavior is normal or not in e-commerce transactions, an algorithm of anomaly detection based on co-occurrence matrix was presented. It accurately modeled user behavior with using co-occurrence matrix, and established the co-occurrence matrix space to obtain profiles of the normal user behavior through the method of principal component analysis. In the detection phase, it acquired the trading patterns of the user in the audit data which converted to the revised co-occurrence matrix, and then to exactly classify the user behavior as normal or malicious by measuring the distance between the patterns and profile employing the second matrix norm. Compared to several other methods, the experiment results show that the proposed method has a higher performance.

关 键 词：异常检测 用户行为 电子商务 共生矩阵 PCA 

分 类 号：TN943.6[电子电信—信号与信息处理]