基于模糊核聚类和支持向量机的鲁棒协同推荐算法

摘要

该文针对现有推荐算法在面对托攻击时鲁棒性不高的问题,提出一种基于模糊核聚类和支持向量机的鲁棒推荐算法.首先,根据攻击概貌间高度相关的特性,利用模糊核聚类方法在高维特征空间对用户概貌进行聚类,实现攻击概貌的第1阶段检测.然后,利用支持向量机分类器对含有攻击概貌的聚类进行分类,实现攻击概貌的第2阶段检测.最后,基于攻击概貌检测结果,通过构造指示函数排除攻击概貌在推荐过程中产生的影响,并引入矩阵分解技术设计相应的鲁棒协同推荐算法.实验结果表明,与现有的基于矩阵分解模型的推荐算法相比,所提算法不但具有很好的鲁棒性,而且准确性也有提高.%The existing collaborative recommendation algorithms have low robustness against shilling attacks. To solve this problem, a robust collaborative recommendation algorithm is proposed based on Fuzzy Kernel Clustering (FKC) and Support Vector Machine (SVM). Firstly, according to the high correlation characteristic between attack profiles, the FKC method is used to cluster user profiles in high-dimensional feature space, which is the first stage of the attack profile detection. Then, the SVM classifier is used to classify the cluster including attack profiles, which is the second stage of the attack profile detection. Finally, an indicator function is constructed based on the attack detection results to reduce the influence of attack profiles on the recommendation, and it is combined with the matrix factorization technology to devise the corresponding robust collaborative recommendation algorithm. Experimental results show that the proposed algorithm outperforms the existing methods in terms of both recommendation accuracy and robustness.