基于仿生模式识别的未知推荐攻击检测∗

摘要

The existing detection approaches can not detect unknown recommendation attacks effectively. Aiming at this problem, an approach based on bionic pattern recognition is proposed. Firstly, items are partitioned into different windows according to their popularity. The ratings given by users for the items in the windows are regarded as the occurrences of random events. Further, information entropy is used to extract features of rating distribution as genuine features for the detection of recommendation attacks. In addition, the technique of bionic pattern recognition is used to cover the samples of genuine profiles in the feature space. Test data outside the coverage are judged as recommendation attacks. The experimental results on the MovieLens dataset show that the proposed approach has high hit ratio and low false alarm ratio when detecting unknown recommendation attacks.%针对已有检测方法不能有效地检测未知推荐攻击的问题，提出了一种基于仿生模式识别(bionic pattern recognition)的检测方法。首先，依据项目流行度划分项目到不同的窗口，把用户对窗口内项目的评分视为随机事件发生。在此基础上，利用信息熵(information entropy)提取评分分布特征作为检测推荐攻击的通用特征。然后，在特征空间中，利用仿生模式识别技术覆盖真实概貌样本，将覆盖范围外的测试数据判为推荐攻击。在MovieLens数据集上进行实验，结果表明，该方法在检测未知推荐攻击时具有较高的命中率和较低的误报率。