基于CNN和犹豫模糊决策的欺诈攻击检测

推荐系统能够有效缓解互联网的迅猛发展带来的信息过载问题，但欺诈攻击的存在制约了推荐系统的健康发展，因此如何准确、高效地检测欺诈攻击是推荐系统安全领域的重要问题。针对传统检测方法依赖专家知识人工提取特征的局限性以及已有基于深度学习的欺诈攻击检测方法存在的硬分类问题，利用卷积神经网络（CNN）自动获取用户空间和时间上的低维表示向量，提出了一种基于CNN和犹豫模糊集的欺诈攻击检测方法CNN-HFS。首先对每个用户分别从评分值、评分偏好和评分时间这三个视角抽取三个行为矩阵，利用双三次插值法对三个矩阵进行缩放得到对应的密集评分矩阵、密集偏好矩阵和密集时间矩阵；然后，将每个用户任意视角下的缩放矩阵视为一个图像，在三个不同视角下分别训练CNN，计算任意用户在每个视角下属于攻击用户类的隶属度；最后，引入模糊犹豫集对多视角下的检测结果进行综合决策，根据决策结果识别出攻击用户。实验结果表明，CNN-HFS在MovieLens 1M数据集上的F1值超过95%，在Amazon数据集上的F1值达到85%。与七种对比方法相比，CNN-HFS在两个数据集上均具有更高的检测精度、召回率及F1值。

Shilling Attacks Detection Based on CNN and Hesitant Fuzzy Sets

Recommender systems can effectively alleviate the problem of information overload caused by the rapid development of the Internet. However, the occurrence of shilling attacks restricts the healthy development of recommender systems. Therefore, how to detect shilling attacks accurately and efficiently is an important problem in the field of recommender systems security. The traditional detection methods rely on hand-crafted features, while the existing detection methods based on deep learning usually have the hard classification problem. In this paper, a novel detection method based on CNN and hesitant fuzzy set is proposed, which can automatically learn the low dimensional representation vectors of users in time and space. Firstly, for each user, three behavior matrices are extracted from the perspectives of rating, preference and rating time, respectively. To reduce the influence of data sparse, matrices are scaled by using bicubic interpolation method to obtain the corresponding dense rating matrix, dense preference matrix and dense time matrix. Next, each scaling matric of users are regarded as an image, and three different CNN classifiers are trained based on these scaling matrixes in three different views respectively. For each user, three membership degrees to the classifier of attack users can be calculated. Finally, the fuzzy hesitant set is introduced to make a comprehensive decision, and the attack users can be identified according to the decision results. The experimental results show that the F1-measure of CNN-HFS is more than 95% on MovieLens 1M dataset and 85% on Amazon dataset. Compared with the seven baseline methods, CNN-HFS has better detection precision, recall and F1-measure on both datasets.