Abstract: The existing unsupervised detection algorithms have a high misjudgment rate for normal user. To solve this problem, a method for detecting shilling attack based on nonnegative matrix factorization is proposed. Firstly, features of the user are extracted from the nonnegative matrix factorization technique. Secondly, the K-means clustering method is used to extract the initial normal user set and the initial shilling attack set. Finally, using features of initial normal user set classifies the initial shilling attack set, to detect shilling attacks. Experimental results show that this algorithms are more effective in detecting the attacks compared to other algorithms.

Key words: recommender systems, nonnegative matrix factorization, shilling attack, detection algorithm

摘要： 针对现有的无监督检测算法对正常用户误检率较高的问题，提出了一种基于矩阵分解的托攻击检测算法。对评分矩阵采用非负矩阵分解技术提取出用户的特征。采用K-means聚类方法对提取出的用户特征聚类，得到初始正常用户集和初始托用户集。利用初始正常用户集的特征对初始托用户集进行二次分类，进一步提高托攻击用户检测的准确率。实验结果表明，所提出的检测算法与其他检测算法相比较能够更有效地检测出托攻击。

关键词: 推荐系统, 非负矩阵分解, 托攻击, 检测算法