基于多目标项目检索的无监督用户概貌攻击检测算法

An Unsupervised Algorithm for Detecting User Profile Attack Based on Multi-target Items Retrieval

为了解决现有的用户概貌攻击检测算法不能对多个受攻击项目同时进行检测的问题,提出一种基于多目标项目检索的无监督用户概貌攻击检测算法.首先,利用双聚类中的Hv-score度量方法,得到对攻击检测有价值的用户概貌集合;然后,在该集合上检索可疑的目标项目,动态生成目标项目树;最后,根据项目的联合评分偏离度,确定受攻击的目标项目并检测出相应的攻击概貌.实验结果表明,该算法无论是检测目标项目还是攻击概貌,均具有较高的精确度.

To solve the problem that the existing user profile attack detection algorithms can not successfully detect multiple items un- der attack at the same time, an unsupervised algorithm for detection of user profile attack based on multi-target items retrieval is pro- posed. We first get the valuable user profile sets for attack detection through the Hv-score metric of biclustering algorithm. Then we retrieval the suspicious items in the sets and dynamically generate the target item tree. Finally, we identify the target items under at- tack and their corresponding attack profiles according to the coalitional rating deviation of the items. The experimental results show that the proposed algorithm has higher accuracy in detecting both target items and attack profiles.